Full text of "microCornucopia :: Micro Cornucopia #53 May90"

See other formats

No. 53

May 1990

$3.95

THE MICRO TECHNICAL JOURNAL

j^TCRO (TJRNUCOPIA

Micro
Controllers

In this, our hysteric final issue, we cover
one of our favorite topics, micro
controllers and their cousins, the data
loggers.

Designing A Micro page 8
Controller

Designing a controller is simple when
you start out with the right micro.

Building A Two-chip page 26
Terminal

The perfect accessory for your micro
controller.

Waylaid By Snakes

A low-down look at objects
(in the grass).

Save The Floppies

Routine, complete with code
recovering bad sectors on floppies.

And More . . .

Shareware /Favorite Tools
Initializing Variables

in Turbo Pascal
Faster Neural Networks

o 74470"19388 I 3

GREAT VALUES FROM MICROSPHERE!

SUPER 25Mhz 386 SYSTEM!

with 4MB RAM and Rotary Voice Coil Hard Drive

For extra speed and reliability we've included a 66MB Mitsubishi
MR535 voice coil hard drive with an access time of 28ms,
and a super fast 1 : 1 interleave HD controller card

25Mhz 80386 CPU, AMI Bios, Full Size MB

4MB of RAM, expands to 8MB on MB

1 .2MB and 1 .44MB Floppy Drives

1 2" Amber Monitor w/Mono Graphics Card

101 -Key Enhanced Keyboard, Focus or Maxiswitch

2 Serial, 1 Parallel, 1 Game, Clock/Calendar

200-Watt Power Supply

Socket for 80287, 80387 or Weitek

Full 1 Year Warranty

Tower Cases w/Power Supply

$2350

(Mega Tower)

w/StdATCase $2250

w/1MB $250 Less

20Mhz386 $100 Less

Standard Tower
$269

AT /386S SYSTEMS

Includes: 1 MB RAM, 1 .2MB & 1 .44MB FD,
40MB Hard Drive (28ms), Fast Hard Drive
Controller, Mono Graphics Video Card,
12" Amber Monitor, 101 Key Keyboard,
Serial(2)/Parallel(1)/Game(1) Ports,
Clock/Calendar, Full 1 Year Warranty.
FREE assembly and testing.

6/10Mhz286/AT $1249

8/12Mhz286/AT 1295

8/16Mhz286/AT 1395

16Mhz386-SX 1449

Video upgrades for our systems
(includes video card and monitor)

ATI Graphics Solution 39

CGA Color 175

EGA Color 349

VGA Color (Analog, 16 bit

w/Amazing monitor) 379

CGA/EGA/VGA (Multisync) 450

FLOPPY DRIVES

360K 69

1.2MB 77

3 1/2" 720K 79

3 1/2" 1.44MB 85

HARD DRIVES

XT20MBMiniscribe/Kalok

8425 (65ms) 229

8425 w/controller 269

XT30MBMiniscribe/Kalok

8438 (65ms) 229

8438 w/controller 269

AT 44MB Mitsubishi MR535,
Rotary Voice Coil (28ms) 429

AT71MBMFM
Voice Coil (28ms) 595

VIDEO CARDS

Mono Graphics w/Parallel port 37

ATI Graphics Solution

Here w/CGA Emulation 79

CGA/EGAA/GA 8-Bit 1 1 9

CGA/EGA/VGA 16-Bit 129

. 'Prices are subject to

The Most Cost Effective Way to Speed Up
your AT or 386 System!

Upgrade your hard drive controller. Discover the NEXT
generation of hard drive controller cards. As the chart below
demonstrates, the speed improvements are incredible!

Controller Card Type

Data Xfer Rate

Standard IBM AT, 3:1 Interleave, MFM 167KB/Sec.

DTK WA2, 2:1 Interleave, MFM 261 KB/Sec.

AT/386 1:1 Interleave, MFM 522 KB/Sec.

Western Digital 1:1, RLL,w/Cache 799 Kb/Sec.

Note: Test results using 10Mhz AT, Mitsubishi MR535 Hard Drive and
SpinRite Disk Optimizer. ^Itt^.

VGA Color ONLY $379... ^St

When you buy a complete Microsphere Computer System. Up-
grade includes: 16 bit VGA video card and 14" Analog VGA
Monitor with a .28 mm dot pitch.

KEYBOARDS

Maxiswitch (AT Only) 101-Key $59

Keytronic* Enhanced 101-Key 55

Focus* 101 -Key Tactile, Switchable

Control/Caps Lock, Dust Cover 59

(#1 Find by Micro C Staff)
* Keyboards are XT/AT switchable

SOFTWARE

MS-DOS 3.3 w/GW Basic 95

DR-DOS 3.3 w/GEM 45

SpinRite II Disk Optimizer

by Gibson Research 79

386Max Memory Manager for

386 Systems 69

CONTROLLER CARDS

AT/386 1 :1 MFM Controller $105

WD 1006-SR2 1:1 RLL w/Cache ....139

CABINETS

XT Slide Case, Lock, LED 35

AT Slide Case, Lock, LED

3 half ht., 1 full ht. drives 59

Baby Tower, w/200 watt PS,

2 half ht., 2 - 3 1/2" Drives,

Baby Motherboard ONLY 129

Mini Tower, w/200 Watt PS,

3 half ht, 2 - 3 1/2" Drives,

holds Full or Baby size MB 1 95

5 Bay Std. Tower, 220 Watt PS,

3 half ht., 1 full ht drives 269

6 Bay Mega Tower, 230 Watt PS. ...229

MONITORS E£

12" Amber Monochrome TTL 88 — ^■^0/1 /7/~V /TTZTOQ

CGA Color RGB 249 -* OU/l//rU / Cl/lOj

cG^vG°A SWitch 362 ° rders only pim^ 1-800-234-8086

Multisync (.31 dot) 489 Tech calls: (503) 388-1194 Hours: Mon-Fri 9:00-5:30

VGA Analog x '

(Amazing 28 dot) 369 855 N.W. WALL • BEND, OREGON 97701

change without notice. Shipping CHARGES will be added. *1 -year warranty/30-day money back (subject to restrictions) j

MOTHERBOARDS

XT/Turbo 4.77/1 OMhz $75

AT6/10Mhz 175

AT8/12Mhz 195

AT8/16Mhz 249

386SX 1 6Mhz w/C&T chip set 359

386 8/20Mhz w/Phoenix Bios,

holds up to 8MB on board 679

386 16/25Mhzw/AMI Bios

holds up to 8MB on board 795

386 33Mhz Call

POWER SUPPLIES

1 50-Watt XT Power Supply 49

200-Watt AT Power Supply 59

230 Watt AT Power Supply 75

EXPANSION CARDS

Clock 18

Game (Joystick) 14

Parallel Port (LPT1, 2 or 3) 18

Serial Port, 2 ports, 1 installed,

(COM1 or 2) 18

2nd Serial Port Kit 18

Serial Port, 4 ports installed 99

Multi Drive Controller, up to

2 drives, Supports 360K,

720K, 1.2MB & 1.44MB 39

3 & 4th Floppy Controller 79

AT/386

AT Multi-IO, 2 Serial, 1 Parallel,

1 Game 35

AT 3MB EMS Memory Card (OK) 99

CHIPS/ETC.

Memory 64K, 256K, 1MB Call

Math Coprocessors Call

XT Components & Systems Call

"HARDWARE MANUFACTURER
SINCE 1983"

Reader Service Number 2

ZORTECH

!MSWI AT&T C+ + RELEASE 2.0
SPECIFICATION

fiMIW MS WINDOWS
COMPATIBILITY

RHSW8 EASY PORTABILITY
FROM MICROSOFT C

(ttl^WS C++DEBUGGER &
EXPANDED C++ TOOLS

MIEWI OS/2 UPGRADE
AVAILABLE NOW!

c** . c»* " ^

c+< roots «

■ - - J /■•

We lisiened carefully to

what you wanted in a next

generation MS DOS C+ +

compiler. The answer is

ZortechC++V2.0

Developer's

Edition.

mmm®

<&** H®8HT
iRsfts1lv£.H«a

You wanted

the latest

AT&TV2.0

features

with the

power

offered by

multiple inheritance and

type safe linkage, so here it

is.

You wanted compatibility
with MS WINDOWS, we
added it.

You repeatedly asked for
easier portability from
Microsoft C, we got the
message, and have written
the library functions you
need.

You wanted the world's first
MS DOS C++ source level

DEBUGGER, and now the
wait is over.

You wanted expanded and
improved documentation,
we both
listened
and
delivered.

You

wanted to
be able to
upgrade
to an

OS/2 version compiler
supporting Presentation
Manager, you did not want
it to cost a fortune, so it's
available
for $150.

You want

to look at

the

standard

library

SOURCE

CODE, so

we are

including

it.

Here is our list of highly
recommended C++ books:

C++ Language/Stroustrup $32.25

C+ + Answer Book/Hansen $26.95

C+ + for C Programmers $29.95

C++ Primer /tippman $30.25

Ask about our new C++ Video Tutorial

For many, EMS
programming support, built
into the compiler is
important, so it's in there
too.

You were happy using the
1 8 classes provided in
C++ TOOLS, but we
revised and expanded it
anyway.

You never asked for a free
TSR library to be included,
but we knew you'd love to
use our neat little package,
so we included it free.

You liked our FLASH

GRAPHICS
package
for its
speed, but
wanted a
C+ +
Class
interface,
so we've
written it.

B-flow To Order:

Already own Zortech
C++? Call the order
hotline for details of our
low cost upgrades.

To order Zortech C++
for the first time, just call
the order hotline. We
accept payment by
Mastercard/Visa/COD.

Alternatively, mail the
coupon below with your
check or credit card
details.

ZORTECH INC.,
1 1 65 Massachusetts
Avenue, Arlington, MA
02174, USA
Voice 617-646-6703
Fax 617-643-7969

ZORTECH LTD.,

106-1 08 Powis Street,

London, SE 18 6LU,

ENGLAND.

Voice {44)-] -31 6-7777

Fax (44)-l-316-4138

Co)

/K\(6) (o)//\}ff))(C))

^© a (o)q>(iJJco)

Yes! Please rush me the following C+ + V2.0 items:

Name

Address

City
Visa/MC#

State

Zip

Exp. Date

Tel

Ai\ MicroSoft trademarks are acknowledged.

□ DEVELOPER'S EDITION$450 (Save $200)

□ C++ COMPILER $199.95
a C++ DEBUGGER $149.95

□ C++ TOOLS $149.95

□ LIBRARY SOURCE CODE $149.95

□ COMPILER & LIBRARY SOURCE $299.95
For US orders please add $5.05 shipping

Reader Service Number 70

□ OS/2 COMPILER UPGRADE $149.95

□ C++ VIDEO COURSE $499.95

□ C++ Language /Sfrousfrup $32.25

□ C++ Answer Book/Hansen $26.95

□ C++ for C Programmers/Phol $29.95

□ C++ Primer/Lippman $30.25
Overseas orders at international mail rates.

INALLY A debugging
tool tough enough to
handle the DOS
Nasties.

New Version 2.0

Nasty over- write? No sweat!

Soft- ICE memory range break points help you
track down memory over-write problems
whether you are doing the over- writing or
another program is over-writing you.

Hung program? No problem!

When the system hangs, you now have hope.
With Soft-ICE you can break out of hung
programs no matter how bad the system has
been trashed. And with Soft-ICE's back trace
ranges you can re-play the instructions that led
up to the crash.

Program too large? Not with Soft-ICE!

Soft-ICE runs entirely in extended memory. This means
you can debug even the largest DOS programs. And
since your program runs at the same address whether
Soft-ICE is loaded or not you can find those subtle
bugs that change when the starting address of your
code changes.

System debugging? Soft-ICE is a natural!

Soft-ICE is ideal for full source level debugging of TSRs,
interrupt service routines, self booting programs, DOS
loadable device drivers, real-time kernels, non-DOS O/Ss
and ROMs. Soft-ICE can even debug within DOS & BIOS.

How Soft-ICE Works

Soft-ICE uses the power of the 80386 to sur-
round your program in a virtual machine.
This gives you complete control of the
DOS environment, while Soft-ICE runs
safely in protected mode. Soft-ICE uses the
80386 to provide real-time break points on
memory locations, memory ranges, execution,
I/O ports, hardware & software interrupts.
With Soft-ICE you get all the speed and power of
a hardware -assisted debugger at a software price.

Don't want to switch debuggers?
You don't have to!

Soft-ICE can run stand-alone or it can add its
powerful break points to the debugger you already
use. Use your favorite debugger until you require
Soft-ICE. Simply pop up the Soft-ICE window to set
powerful real-time break points. When a break point
is reached, your debugger will be activated automatically.

MagicCV with Soft-ICE

Using Soft-ICE with CodeView gives you the features
necessary for professional level systems debugging.
MagicCV and Soft-ICE can work in concert with Code-
View to provide the most powerful debugging platform
you will find anywhere.

"These may be the only two products I've seen in
the last two or three years that exceeded my wildest
expectations for power, compatibility and ease-of-use."

— Paul Mace
Paul Mace Software

Soft-ICE $386

MagicCV $199

MagicCV for Windows $199

Buy Soft-ICE & MagicCV(W)

-Save $86.
Buy MagicCV and MagicCVW

-Save $100.
Buy All 3 -Save $186.

30 day money -back guarantee
Visa, MasterCard and
AmEx accepted

New Soft-ICE 2.0 features

O Back Trace Ranges

d Symbolic & Source level debugging

O EMS 4.0 support with special EMS

debugging commands
n Windowed user interface

Nu-Me ga

TECHNOLOGIES

CALL TODAY (603) 888-2386
or FAX (603) 888-2465

RUN CODEVIEW
IN8K

WiagicCV

CodeView is a great integrated debugger, but
it uses over 200K of conventional memory.
MagicCV uses advanced features of the
80386 to load CodeView and symbols in
extended memory. This allows MagicCV to
run CodeView in less than 8K of conven-
tional memory on your 80386 PC.

NEW VERSION FOR C 6.0

Attention Windows Developers!

Version available for CVW.

P.O. BOX 7607 ■ NASHUA, NH ■ 03060-7607

2 MICRO CORNUCOPIA, #53, May, 1990

Reader Service Number 110

THE

MICRO

TECHNICAL JOURNAL

jyjICRO fORNUCOPI A

MAY 1990— ISSUE NO. 53

20
26

46
54
70

H. Ward Silver

Designing A Microcontroller

Ward puts together a very simple data logger
and microcontroller based on the 68HC11.

Bruce Eckel
Waylaid By Snakes

A fun look at C++ if you don't object to serpents.

Karl Lunt

Building A Two-chip Terminal

Here's a simple, low-power terminal with a two-line by twenty-column
display.

Lance Dannon Bresee

A Roundoff Roundup

Tired of round-off errors? This should help your calculations.

Dave Gwillim

Initializing Variables In Turbo Pascal

An elegant fix for a popular compiler.

D. Gilbert Lee

Fast Neural Networks

Larry Fogg

Save The Floppies

Stuart R. Bell

Build A Composite Video Adaptor

57
64
65
81

86 World

Culture Corner

On Your Own

Units and Modules

OO Shareware
90 Techtips

85 Tidbits
96 Last Page

©§MI3

Cover Illustration by* Greg Cross.

MICRO CORNUCOPIA, #53, May, 1990 3

THE MICRO TECHNICAL JOURNA.L

MICRO CORNUCOPIA

Editor and Publisher

David ]. Thompson

Associate Editors

Gary Entsminger
Larry Fogg
Cary Gatton

Contributing Writers

Bruce Eckel

Michael S. Hunt

Alfred E. Newman

Karl hunt

Accounting

Sandy Thompson

Advertising & Distribution
Reader Services

Nancy Ellen Locke & Laura Shaw

Graphic Design & Production

Carol Steffy

MICRO CORNUCOPIA (ISSN 0747-587X) was
published bi-monthly for $18 per year by Micro
Cornucopia, Inc. 155 NW Hawthorne, Bend, OR
97701. Second-class postage paid at Bend, OR and
additional mailing offices. POSTMASTER: Send
address changes to MICRO CORNUCOPIA, PO
Box 223, Bend, OR 97709. ^ Recycled Paper

SUBSCRIPTION RATES:

1 yr. (6 issues)

$18.00

2 yr. (12 issues)

$34.00

3 yr. (18 issues)

$48.00

1 yr. Canada & Mexico

$26.00

1 yr. Other foreign (surface)

$36.00

1 yr. Foreign (airmail)

$50.00

Make orders payable in US. funds on a US. bank

CHANGE OF ADDRESS:

Please send your old label and new address to:

MICRO CORNUCOPIA

P.O. Box 223
Bend, Oregon 97709

READER SERVICES:

For orders and subscription problems call (503)
382-5060, 9 am to 5 pm, Pacific time, M-F. FAX
your VISA or MC order to us, our FAX number is
(503) 389-6833.

BBS - 24 hrs. 300-1200-2400 baud
8 Bits, No Parity, 1 Stop Bit 503-382-7643

ISSN 0747-587X

The

Audit

Bureau

AROUND

THE BEND

By David J. Thompson

Last Hoorah

4 MICRO CORNUCOPIA, #53, May, 1990

I'm not sure how to begin. I've started this editorial
so many times, in my daydreams, in my sleep, in the
shower.... But nothing I've come up with really feels
like I feel.

I'm closing down Micro C and I don't know what I'll
be doing next. (Wait till mom reads this.... "Dad! Our
little David's quitting his job. Quick, clean up the spare
room.")

You may be wondering why it's closing rather than
moving on to other ownership. Good question. Three
groups have come forward, interested in taking on
Micro C. After long discussions, none felt they could put
together both the finances and the talent to make it fly.

When a magazine folds (a strangely appropriate
word), other magazines often pick up the subscribers.
As far as I know, the only subscribers no one wanted
were the ones taking Profiles (aimed at Kaypro owners)
and CP/M Review. So we've had offers from several ma-
gazines wanting to finish your subscriptions. How did I
choose?

Some were willing to pay for your names. Some
weren't. Some were already popular with Micro C sub-
scribers. Some weren't. Some were very interested in
picking up the articles already in the pipe for future is-
, sues. One even wanted to continue "Around The Bend"
in its entirety. (Led to an, uhm, lengthy discussion
about sanity.)

After considering all the options, I selected Computer
Language. To finish out your Micro C subscription, you'll
get your choice of Computer Language, AI Expert, or
Embedded Systems Programming. If you already sub-
scribe (25% of you already take Computer Language),
they'll extend your subscription. Just fill out the form
you'll receive in April.

If you've got a year of Micro C coming, you'll get a
year of Computer Language — and you'll be getting it at a
really good discount (CL's normally $29 a year, Micro
C's just $18). Plus, Computer Language's standard
policy is: if you decide you don't like their magazine,
they'll send you a refund for all remaining issues. All
you have to do is let them know.

Look, I know that Computer Language isn't Micro C.
When I looked closely the past few months, I was sur-
prised how far we'd moved away from the other com-

Continued on page 74

Tr v /, Or S7< J? Tr?\ < l7TrX\ O

IDteiMJ ¥« UsdMi ITIm

^~^ rying to modify a program without its
source code is tough enough without
having to use an inferior disassembler.

That's why you should look at DASM™
before you invest in any disassembler
product.

Most disassemblers complicate the
process with errors: giving you incorrectly
placed labels, incorrect and misplaced code
and data, data treated as code, or code
treated as data.

DASM™ makes the modification process
easy by keeping track of segment register
usage and mapping code in its execution
sequence. It determines which parts of the
file are code, which are data, and what
labels exist. You simply edit the assembly
language output, make the necessary
changes to the program, and reassemble.

Japan's #1 Seller - Made in America

DASM's simplicity, accuracy and power
have made it the top-selling disassembler
in Japan for two straight years. The
reason - DASM™ does what other dis-
assemblers can't:

a Handle problematic segment and instruc
tion address register loads through interac-
tive mode

□ Track segment register usage

□ Generate appropriate ASSUME's and
segment maps

□ Handle multiple entry points

□ Track memory references

• Direct

• Via register

• Implied in OS function requests

□ Handle transfer vectors
n Handle .EXE, .COM and .BIN files
upto200K

Try DASM™ Risk-Free!

DASM™ is available through the Pro-
grammer's Shop for only $225. And it's guar-
anteed to outperform any disassembler on
the market.

Order your copy now and throw your
worst disassembly problem at it. We think
you'll be impressed. If not, just return
DASM™ for a full refund.

Why settle for anything less?

Software
To Order Call
301-752-1348

For more information, call (301) 752-1348.

To demonstrate DASM's superior perform-
ance, we gave the same disassembling task
tobothDASM™andSourcer™(Version2.18).

Compare
For Yonnf self

While DASM™ generated the correct
assembly code, Sourcer made at least six
major errors.

m®m

SOURCER™

(0) Sourcer™ has incorrectly Assumed
DS to the segment address of the code.
(DS is actually pointing to the PSP on
entry.)

© Sourcer™ continues to have the
incorrect Assume for the DS, even after
being loaded with a segment value refer-
enced in the program header.

© Sourcer™ does not reference the seg-
ment value symbolically, so any changes
to the program which alter its location
will not be reflected.

© Since Sourcer™ generates no seg-
ment labels relative to the DS segment,
changes in the DS segment will not be
reflected.

In short, if you're not using DASM™, be
prepared to spend a long time editing the
file to add and fix the missing and incor-
rect labels - and expect a lot of bugs!

Trademark/Owner: DASM™/]BSoftware;
Sourcer™/V Communications.

dataJe
dataJe

data_4
loc_1:

equ
equ

segment para public

assume cs:segj , dsiseg^a -^(SD"

6F15h © iyiff$

SEGMENT

LABEL BYTE

db 'Print string message$'

LABEL BYTE

db 'P'

org 20H

ENDS

END C1F15H

Reader Service Number 213

MICRO CORNUCOPIA, #53, May, 1990 5

Rack & Desk
PC/AT Chassis

Integrand's new Chassis/System is not
another IBM mechanical and electrical
clone. An entirely fresh packaging design
approach has been taken using modular
construction. At present, over 40 optional
stock modules allow you to customize our
standard chassis to nearly any requirement.
Integrand offers high quality, advanced
design hardware along with applications
and technical support all at prices competi-
tive with imports. Why settle for less?

Rack & Desk Models

Accepts PC, XT, AT Motherboards
and Passive Backplanes

Doesn't Look Like IBM

Rugged, Modular Construction
Excellent Air Flow & Cooling

Optional Card Cage Fan

Designed to meet FCC

204 Watt Supply, UL Recognized
145W & 85W also available

Reasonably Priced

Passive
Backplanes

*.i i f.f s-ssmmM !

RESEARCHCORR

Call or write for descriptive brochure and prices:
8620 Roosevelt Ave. • Visalia, CA 93291

209/651-1203

TELEX 5106012830 (INTEGRAND UD)

FAX 209/651-1353

We accept Bank Americard/VISA and MasterCard

IBM, PC, XT, AT trademarks of International Business Machines.
Drives and computer boards not included.

Letters

So Long

Today I heard that Micro Cornucopia
is to be no more. This greatly saddens
me even as I recognize the need for
change in all things. Your magazine
filled a niche in the journalistic world
that was very special and unique. You
represented all of us plain folks, who
did not hide behind the shirt and tie but
who loved working with computers at
every level.

You spoke to us, as one enthusiast to
another, not using technojargon to hide
a lack of real understanding so common
in other journals. You held gatherings
where we laughed and joked with the
literary mentors who taught us and
guided us on our own paths to knowl-
edge, even while we wiped the bar-b-
que sauce off our lips.

You became family. You will be
missed. My best wishes and thoughts go
out to each and every one of you at
Micro C who have touched my life.
Godspeed And Thank You.

Al J. Szymanski (Big Al)
8991 Edcliff Ct. SE
Aumsville, OR 97325

Editor's note: Thanks AL The entire staff
remembers, only too vividly, your smashing
presence during SOG volleyball. Godspeed
to you, also.

Bend Memories

I have just sent in my subscription re-
newal. Late. I can't believe that just be-
cause I forgot to renew, you stopped
sending me the magazine! Didn't you
know that I still wanted it?

Seriously, I think Micro C is the best
computer mag available. You can im-
agine my shock when I first saw it a
couple years ago. I couldn't believe such
a great mag was published in Bend, of

Reader Service Number 22
6 MICRO CORNUCOPIA, #53, May, 1990

all places. I mean, I used to live in Bend
when I was 14, in the year 19-mumble-
mumble. I don't suppose Bend has
changed any since then? As I recall,
Bend was very small.

I say, Bend was very small. (Did you
say, "How small was it?") It was so
small that if you wanted a business
phone number, you looked it up on the
yellow page. It was so small that every
time I turned on my radio, the electric
trolly slowed down. It was so small we
didn't have a regular village idiot; we all
took turns. Bend had a newspaper,
though. ("Really?") Yeah, a traveling
salesman left it at the feed store.

I may be exaggerating slightly, but
the most interesting thing I did all year
there was ride my bike up the dead vol-
cano outside town. I had just seen The
Last Days of Pompeii and, being 14, 1 had
hopes it might erupt again and liven
things up a bit. No such luck.

Sounds like I missed something by

not attending the Rocky Mt. SOG. But I

had been to Gunnison before, and

having recently seen Deliverance, I

Letters continued on page 68

B"****

Qood LucIq

'Dave & Sandy
Jennifer & ( Erin

The $25

Network

Try the 1st truly low cost LAN

• Connect 2 or 3 PCs, XTs, ATs

• Uses serial ports and 5 wire cable

• Runs at 115 K baud

• Runs in background, totally transparent

• Share any device, any file

• Needs only 14K of ram

Skeptical? We make believers!

CW 1 5 , 000 sM uWiWk
Information Modes

P.O. Drawer F
Denton, TX 76202
817-387-3339 or 1-800-628-7992

Why waste money on simple file transfer systems?

1 \S

Love,
(Don & %m
Conoid & Courttanct

Reader Service Number 149

.y*

,\ \

!x-*

i \

,.! H

MICRO CORNUCOPIA, #53, May, 1990 7

Designing A Microcontroller

The MC68HC1 1 In Action

Ward walks us through a microcontroller
design project based on the 6811. If you're
thinking about designing a monitor to watch
the real world or control a process this
project, the DataBear, should make great
reading. (Now that I've spilled the name of
his board, you suspect I'm going to make one
of my unbearable puns. Right?)

The best way to describe a particular
microprocessor is to show how it's
used. In this article, I'm using the
Motorola MC68HC11 in a battery-
powered data logger, the DataBear from
Langan Products, Inc., of San Francisco.

The 6811 is one of the more flexible
and powerful eight-bit microcontrollers.

8 MICRO CORNUCOPIA, #53, May, 1990

See Figure 1 for a list of features and Fig-
ure 2 for the way it's organized.

Not only is the 6811 well supported
by software and hardware development
tools (including C compilers and emula-
tors), it's quite cheap. Unlike some of the
latest whiz-bang processors, a 52-pin
version with 512 bytes of EEPROM costs
only about $12 in onesies. Plus, you can
easily interface inexpensive 8-bit support
chips to the 6811.

See Figure 3 for a list of variations on
the 6811.

What Is A DataBear?

I designed the DataBear as a small,
simple unit, powered by just about any-
thing and capable of monitoring ordi-
nary analog data at very ordinary rates.

Of course, I wanted to include as
many features of more expensive log-
gers as I could while keeping down the
chip count and printed-circuit size. The
6811 makes that possible.

The DataBear packs the 6811, 8K or
32K of static RAM, 8K or 32K of
EPROM, bus logic, real-time clock, serial
interface, temperature sensor, and two
external channels of analog input onto a
board 2.6x3.0 inches. With the battery
packs, the whole unit is only 1x4x6 in-
ches. It's portable and powerful, thanks
to the 6811.

Slow And Steady Continuous Mode

I set up the DataBear to collect data
over a long period of time, from once
per second to once every 18 hours. Each

By H. Ward Silver

RBR Design

P.O. Box 1608

Vashon,WA 98070

(206)463-2833/9173

Fax (206) 463-9122

Figure 1 — Basic 6811 Features

Internal Memory: 256 bytes static RAM

512 bytes EEPROM

- 12K bytes mask ROM (programmed at factory)
External Memory: 64K byte address space with mux'd bus
Serial Communications Interface (SCI) , to 131K baud
Serial Peripheral Interface (SPI) , clock speed to 1/2 crystal freq.
Parallel I/O: up to 38 bits, some dedicated Input or Output
16-bit Timers: Four-Stage Prescalar

3 Event Capture Counters (ICF)

5 Counter Controlled Outputs (OCF)

Pulse Accumulator Input (PAI)
8 Channels of 8-bit Ratiometric A/D Conversion

2 Hardware External Interrupts: XIRQ (non-maskable) and IRQ (maskable)
Programmable Real-Time Interrupt and Watchdog Timer
WAIT/STOP Low Power Modes
6801-based instruction set plus: 16x16 Integer and Fractional Divide

Bit Manipulation Instructions
Direct, Extended, and Indexed (2 16-bit Index Registers) Addressing

sample is an average of eight rapid
samples, thus minimizing high
frequency noise.

You'll probably use Continuous
Mode for most field, shipping, or remote
applications. When you're sampling at a
very slow rate, the processor can shut
down between samples, saving power.

Programmable Fixed Mode

A second mode uses the real-time
clock to take bursts of data at preset
times. This is called Fixed Mode and it
lets you control the time resolution of
the data.

Real Time Mode

The third logging mode samples on
request by a host computer. A command
string from the host triggers the sam-
pling. The Bear then reports its data in
either binary or ASCII. The maximum
rate is approximately 20 samples per
second, limited mainly by the speed of
the serial link.

The DataBear's time-controlled
operation makes heavy use of the 6811's
SPI port to service clock interrupts and
read the clock time and date. Because of
the SPI's high speed, these transactions
don't take long, providing for low
power consumption and a high sam-
pling rate.

Analog Conversion Numbers

The DataBear has three analog in-
puts. The 6811 has an eight-channel
A/D Converter, but I used two to moni-
tor the power and left three for expan-
sion. Of the three left, I used one channel
to monitor internal temperature. You get
the other two for data. The converter res-
olution is eight bits, its accuracy limited
by the voltage reference and by the per-
formance of your sensor.

Eight bits of resolution is acceptable
resolution for the vast majority of field
data logging requirements. For example,

MICRO CORNUCOPIA, #53, May, 1990 9

with the built-in LM34 temperature sen-
sor, we measure to 128 degrees
Fahrenheit with a resolution of 1 degree.

Host Interface

Interaction with the DataBear hap-
pens over an RS-232 serial port at 9600
baud. Any type of computer can talk to a
DataBear. To avoid the bulky and expen-
sive DB-25 type connector, I used a tele-
phone handset (RJ-11) style jack.

The serial port uses only three lines:
receive data, transmit data, and ground.
The modular jack has proven to be a
wise choice because of its durability. I
chose a handset-size jack to prevent ac-
cidental connection to a 50 Volt phone
line with its 100 VAC ringing signals.
These voltages would damage the Bear.

Processor & Memory

See Figure 4 (on pages 14-15) for the
functional schematic. If you're familiar
with the Motorola 6801 family, you'll see
that I've implemented a standard multi-
plexed memory bus. The data lines and
the eight least-significant address lines
are multiplexed at the processor, and
demultiplexed by a 74373 octal latch.
The 6811 Address Strobe (AS) signal
controls the '373.

The 6264 8Kx8 static RAM and 27C64
8Kx8 EPROM are the only bussed dev-
ices. The RAM is mapped at $0000 and
the EPROM at $E000. Because of the
simplicity of the external memory-map,
you can use A15 as the master chip
select, gated with the processor's E clock
for timing. The 6811's on-chip resources
reduce or eliminate the need for separate
peripheral chips such as UARTs, parallel
I/O, timer/counters, etc.

As memory prices fall, I'll change to
32Kx8 RAM. This way the Bear will
handle more data and more channels.
Adding more EPROM space for en-
hanced firmware is easy also. Because of
the size of the printed-circuit board,
adding more complex memory mapping
would require a CMOS PAL.

A neat feature of the 6811 is that the
internal RAM (256 bytes) and the inter-
nal I/O control registers (64 bytes) can
be "moved" by firmware to any 4K
boundary. These can be overlaid with
external memory without bus conflicts.
Both the internal RAM and control regis-
ters can overlay each other, sacrificing
the lowest 64 bytes of internal RAM.

A less-than-neat feature of the 6801-
family multiplexed bus timing, often
overlooked by designers, is that data is

Figure 2 — Organization of 6811

MODA MODB
(ITR) (V STBY )

MODE CONTROL

osc. |

CLOCK LOGIC |

(V PP ) WW RESET

INTERRUPT LOGIC

ROM 8K BYTES

TIMER
SYSTEM

EEPROM 512 BYTES

RAM 256 BYTES

BUS EXPANSION
ADDRESS

ADDRESS/DATA ^ t

FT rr

STROBE AND HANDSHAKE
PARALLEL I/O

TTT

t t ♦ I ! i'

.,.,., ., .. ■

SERIAL
PERIPHERAL
. INTERFACE

SERIAL

COMMUNICATION

INTERFACE

,7J fl ., ,

v D d
pss

PORT D

■■ '

"Vrh

"Vrl

A-D CONVERTER 1

^3S

NOT BONDED ON
48-PIN VERSIONS

A — Block Diagram

| 7

S X H I N Z V C

8-BIT ACCUMULATORS A AND B

OR 16-BIT DOUBLE ACCUMULATOR D

INDEX REGISTER X

INDEX REGISTER Y

STACK POINTER

PROGRAM COUNTER

CONDITION CODE REGISTER

CARRY/BORROW FROM MSB

■ OVERFLOW
ZERO

■ NEGATIVE

■ l-INTERRUPT MASK

HALF CARRY (FROM BIT 3)

■ X-INTERRUPT MASK

■ STOP DISABLE

B — Programming Model

10 MICRO CORNUCOPIA, #53, May, 1990

lail'gWHSTOLl.C I —

while (TRUE) { /• proces 1B3BMS3DM|

i 1 natch or error •/

j-getseqtkeybui', i ); 1 Ij^BiikW^^^^B

ny length key-sequence »/

u - tn i H ■

rst keypress «/

liffffll

/« Check Tor delinite Keystroke tlacros

if (»keybuf ■■■ dei irt it I

Add flacro
Display Bacros

/* Check if displaydbi Load fron Disk

Save to Disk

if iCkeybuf >; ' ') a 1
printf ("xc",*keybtif );
•codbuf** - »keybuf;
»codbifjJ» - 00;
•codbtif -■ KFF;
return(TRUE);

High byte 00 for chars

■"""" IHSIftLl.C(675) : error 65: 'codbif : undefined

^IHDOU H=

Edit source file; then press <CTRL-E> for next error, <ESC> for »emi

Introducing . . .

The 1st Family of

Low Cost, Powerful

Text Editors

VEDIT Jr. $ 29
VEDIT $ 69

VEDIT PLUS $185

Finally, you can choose the best editor for your needs without
compromising performance or paying too much. And organiza-
tions that want the "same" editor for everyone can pick VEDIT®
for most users and VEDIT PLUS for their power users.

The new family of VEDIT text editors are upwards compatible,
easy to use and offer exceptional performance, flexibility and
stunning speed. (3 to 30 times faster than the competition on
large files where speed really counts.)

Call for your free evaluation copy today. See why VEDIT has
been the #1 choice of programmers, writers and engineers
since 1980.

VEDIT Jr.— Unmatched performance for only $29.

All VEDIT editors include a pull-down menu system with "hot
keys," context sensitive on-line help, pop-up status and ASCII
table, a configurable keyboard layout and flexible, unlimited
keystroke macros. Edit files of any size and any line length.
Perform blockoperationsbycharacter, line, fileorcolumn. Undo
up to 1000 keystrokes— keystroke by keystroke, line by line, or
deletion by deletion. Automatic indent, block indent and paren-
theses matching speed program development. Word wrap,
paragraph formatting, justification, centering, adjustable mar-
gins and printing for word processing. Run DOS programs.

VEDIT-A best value at only $69.

Simultaneously edit up to 36 files and split the screen into win-
dows. Search/replace with regular expressions. Includes the
best compiler support available— menu driven, easy selection
of compiler options, supports "Include" files and MAKE utilities.

VEDIT PLUS -Ultimate programmer's tool for only $185.

VEDIT PLUS adds the most powerful macro programming
language of any editor. It eliminates repetitive editing tasks and
permits creating your own editing functions. The macro
language includes testing, branching, looping, user prompts,
keyboard input, string and numeric variables and control over
the size, position and color of windows. Source level macro
debugging with breakpoints and tracing. Macros developed
with VEDIT PLUS also run under VEDIT.

30 day money-back guarantee. Call for pricing of XENIX, OS/2
and FlexOS versions. Very attractive quantity pricing is avail-
able for schools, hardware and software vendors.

VEDIT and CompuView are registered trademarks of CompuView Products, Inc. BRIEF is a
trademark of UnderWare, Inc. Norton Editor is a trademark of Peter Norton Computing Inc. QEdit
is a trademark of SemWare.

•Supports IBM PC, XT, AT, PS/2 and clones with CGA, MGA, EGA, VGA, Wyse 700, Amdek 1280

and other displays. Also supports Concurrent DOS, DESQview, Microsoft Windows,

PC-MOS/386 and most networks.
'Also available for MS-DOS (CRT terminals), Tl Professional and others.
•Free evaluation disk is fully functional and can edit small files. Reader Service Number 7

FREE Evaluation Copy'
Call 1-800-45-VEDIT

Compare Features and Speed

VEDIT

BRIEF 2.10

Norton 1.3

QEdit 2.07

Pull-Down menus

Yes

Pop-Up ASCII table

Yes

Keystroke macros

100 +

Regular Expressions

Yes

"Cut and Paste" buffers

100

Text (book) markers

Undo keystroke by keystroke

Yes

Undo line by line

Yes

Normal/max Undo levels

500/1000

30/300

Variable tab positions

Yes

Configurable keyboard

Yes

Difficult

Integrated mouse support

Yes

FILE LIMITS

Edit files larger memory

Yes

Difficult

Maximum line length

>8096

512

65,535

512

Maximum lines/file

8,388,607

65,535

> 65,535

20,000

COMPILER SUPPORT

None

Menu driven

Yes

Select Compiler options

Difficult

Support "Include" files

Yes

BENCHMARKS 50K FILE

Simple search

0.2 sec

1 sec

0.3 sec

Save and continue

1 sec

2 sec

1 sec

1000 replacements

3 sec

19 sec

17 sec

2.5 sec

BENCHMARKS 3 MEG FILE

Simple search

1:40 min

1:36 min

Cannot

Save and continue

1:05 min

3:23 min

Cannot

60,000 replacements

3:18 min

1:44 hour

Cannot

Block-column copy (40 x 200)

2 sec

30 sec

Cannot

2 sec

Insert 1 Meg file in

middle of 1 Meg file

1:11 min

15:13 min

Cannot

PRICE

$69

$195

$75

$54.95

CompuView

1955 Pauline Blvd., Ann Arbor, Ml 48103
(313) 996-1299, Fax (313) 996-1308

MICRO CORNUCOPIA, #53, May, 1990 11

Figure 3— MC68HC11 Family Members

Device Number

ROM

EEPROM

RAM

CONFIG 3

Comments

MC68HC11A8

512

256

$0F

Family Built Around this Device

MC68HC11A1

512

256

SOD

Same Die as 'A8 but ROM Disabled

MC68HC11A0

256

$0C

Same Die as 'A8 but ROM and EEPROM Disabled

XC68HC11B8

512 1

256

$0F

Early Experimental Version

XC68HC11B1

512 1

256

SOD

'B8 with ROM Disabled

XC68HC11B0

256

soc

'B8 with ROM and EEPROM Disabled

MC68HC11E9

12K

512

$0F

Four Input Captures and Bigger RAM and 12K ROM^

MC68HC11E1

512'

512

SOD

'E9 with ROM Disabled 4

MC68HC11E0

512

SOC

'E9 with ROM and EEPROM Disabled 4

MC68HC811A2

2K 2

256

$FF

No ROM Part for Expanded Systems

MC68HC11D3

192

N/A

Economy Version, No A-D and Smaller Memories

Notes:

1. The EEPROM on B Series parts requires an external 19- volt supply for programming and is not byte erasable.

2. This 2K EEPROM is relocatable to the top of any 4K memory page. Relocation is done with four bits in the CONFIG register.

3. CONFIG register values in this table reflect the value programmed prior to shipment from Motorola.

4. Available in 1988.

only guaranteed for 10 nanoseconds fol-
lowing the trailing E-clock edge. Either
you carefully watch the number of gate
delays between the processor E-clock
and data receiving device or you
generate an intermediate clock edge.
6809 users have the quadrature (Q) clock
edge available to help grab data.

You can also operate the 6811 in the
Single-Chip Mode if you install the pro-
gram in the internal masked ROM (great
for high- volume developers).

For the best of both worlds, Motorola
makes a companion chip, the 68HC24,
which restores lots of parallel I/O to an
external memory design. This device
connects to the 6811's address/data bus
and restores the 16 bits of parallel I/O
lost to the bus. It's as if the parallel ports
were built into the 6811! So you can have
it all.... For a price.

SCI Software

The 6811 has an on-chip SCI for serial
communications. The SCI can generate
interrupts or set status flags and it
senses Framing and Overrun errors.

Though the SCI won't automatically
generate and check parity, it will send
and receive parity bits supplied by soft-
ware. Plus, you can use incoming data to
wake up the processor after you've left it
in WAIT or STOP (low-power) mode.

Since baud rates get generated by
dividing the processor's clock, you'll
want to choose the processor frequency
carefully. I selected 8.0 MHz so the Bear
can talk at all the standard baud rates.

The 681 l's SCI output is logic-level,
not the ±12V standard for RS-232. I
chose not to use the familiar 1488/89 RS-
232 interface because of their hefty
power needs and their use of ±12V.

Instead I used the Maxim MAX232. It
has two line receivers and two line
drivers and it has a charge pump for
generating RS-232 line-output voltages.
By controlling the amount of power to
the charge pump, I control power con-
sumption.

Real-Time Clock And Serial
Peripheral Interface (SPI) System

During the initial development of the

DataBear, I used the National MM58167
clock chip. This venerable device
worked well enough, but it didn't offer
absolute year clocking. And it required
four I/O lines plus control lines. On a
2.5x3.0 inch board, that's expensive!

Then I found the RCA 68HC68T1
clock family, now second sourced by
Motorola. The '68T1 family offers a 6811-
compatible SPI port, absolute year,
weekday, time to hundredths of a sec-
ond, and 32 bytes of static RAM in some
family members. I selected a 32 KHz
tuning fork watch crystal as the time-
base, although you can use crystals up to
8 MHz with higher power consumption.

A caveat for clock circuit designers:
tuning fork watch crystals have a poor
temperature coefficient compared to the
standard HC-6/U quartz plate crystals.
A typical watch crystal can be adjusted
to 32,768.0 Hz ±0.1 Hz. However, even
at that, the clock can gain or lose several
seconds per week.

It's funny; tell people that you have 3
ppm accuracy and they love it, but
several seconds per week is awful! How

12 MICRO CORNUCOPIA, #53, May, 1990

do watches keep time so well? It helps to
keep the crystal at a constant 90 degrees
by strapping it to your wrist! (A warm
hand means a timely person.)

The DataBear communicates with the
clock via the 6811 SPI port. The SPI sys-
tem is a high bit-rate, synchronous,
Master /Slave, serial data port using two
data lines and a clock.

The interface supports full duplex,
multiple masters, handles limited data
collision detection, and even offers inter-
rupts and status flags.

In most simple single-processor sys-
tems, the processor gets to be bus
Master. All other SPI devices become
Slaves. You might scratch your head try-
ing to work out clock polarity (CPOL)
and phase (CPHA), but careful scrutiny
of the data sheets usually solves the
problem. With only four possibilities,
you can always psyche it out with a
scope or debug monitor.

Once you've configured the SPI, all
you need is software which knows how
to talk with each slave. Each slave will
probably speak a slightly different pro-
tocol.

Our 68HC68T1 clock speaks data in
nybbles and (of course) in a reverse

order, but that's about par for the
course.

The Master sends data serially out the
Master Out Slave In (MOSI) pin, while
simultaneously generating a shift clock
on the SCK pin. Slaves send data to the
Master on the Master In Slave Out
(MISO) pin. Tie the 681 l's SS line high
when it's the Master Device; use the pin
as a device select (Slave Select) when it's
the Slave.

Make sure that the SPI clock rate is
not too fast for any of the Slave devices,
especially CMOS parts. Each Slave will
probably have its own device select line,
so you can program a custom clock rate,
polarity and phase for each device.

I've found the SPI is probably the
most trouble-free subsystem in the entire
DataBear. More SPI parts seem to be
available each month. Other serial dev-
ices, such as A/D or D/A Converters,
often work fine even though they are not
marketed as SPI compatible.

A/D Converter Subsystem

The 6811 A/D Converter is a ra-
tiometric converter. The output of the
converter is a digital ratio of the input
signal to the voltage references. With the

6811, these references are V r i and V r h.
Any voltage between V r i and V r h will re-
sult in an eight-bit output according to
the formula:

Output (0-255);

Input -V r [

x255

Vrh -Vri

It's easy to tie V r i to ground and V r h
to V cc , but digital noise can be a problem.
You can reduce the noise by adding an
RC-filter to the V cc power at V r h. Better
still, use a zener diode or adjustable volt-
age reference to regulate V r h.

Also be careful about how you route
a ground to V r i. If digital currents flow in
ground traces used for V r i, you'll have a
noise problem again.

I used a National LM235-2.5 as a ref-
erence diode for the DataBear. The
LM136/236/336 series of references are
adjustable, conditioned zener diodes.
The nominal tolerance is 5%, and I use a
pot to divide the reference's output
down to 1.280 Volts. Refer to the "Volt-
age Reference" section of Figure 4 for a
schematic of the DataBear reference cir-
cuit.

The A/D Converter in the 6811 has
different modes of conversion controlled

Continued on page 16

QEdit"

ADVANCED .■-—

The Fast, Easy-To-Use, Full-Featured Text Editor. Now Also Available as a TSR Utility!

Hf ED/T

EDITOR

QEdit has always been the text
editor that combined price, perfor-
mance and value. QEdit TSR makes
that value even better by giving
you instant access to the power of
QEdit— no matter what program
you're working in.

QEdit is packed with features:

□ Completely configurable, includ-
ing keyboard and colors

D Simple to install, requires less than
50K disk space

Q Simultaneous file editing— limited
only by available memory space

□ Open up to eight windows

□ "Pop-Down" menu system and
customizable Help screen

D Easy-to-use macro capability-
including keyboard recording

D Column blocks

D Recover deleted text

D Exit to DOS or DOS shell from

within QEdit

D Execute command-line compilers

from within QEdit

Q Wordwrap and paragraph

reformat capability

D Import files and export blocks

D Great for use with laptops— QEdit

saves battery drain by editing files

entirely in memory

System Requirements

QEdit requires an IBM PS/2, PC/AT, PC/XT,
PC, PC/Jr, or compatible. Minimum system
requirements are 64 KB of memory, PC-DOS
2.0 or MS-DOS 2.0 or greater, 50 KB of disk
space. QEdit runs GREAT on floppy based
systems and laptops.
QEdit TSR also requires EMS, XMS, or a
hard disk.

6 6 QEdit is a
great piece of soft-
ware. Highly
recommended. 9 9
John C. Dvorak,
PC Magazine
September 12, 1989

6 6 QEdit is, without question,
the smallest, fastest, most
versatile text editor in the
DOS world. 9 9

P. L Olympia, Ph.D.,

DBMS

July, 1989

Winner, 1989 Data Based

Advisor Reader's

Choice Award

QEdit"$S495

ADVANCED *J I

ALSO AVAILABLE
QEdit™ for OS/2

UNCONDITIONAL
MONEY BACK GUARANTEE

INTRODUCING QEDIT TSR

Instant Access to the

World's Best Editor

SPECIAL PACKAGE PRICE OF ONLY

$®(p)00 'NCLUDES

^^ QEDIT ADVANCED

To order direct call

404-641-9002

extension 12

'MastafCart

■ Cards

Add $3.00 for shipping— $ 1 0.00

for overseas shipping. UPS 2nd

DAY AIR available within

the U.S. for ONLY $5.00

COD's accepted— please add $3.00
Georgia residents add 4% sales tax

SemWare®

4343 Shallowford Rd. • Suite C-3
Marietta, GA 30062-5003

QEdit is a trademark and SemWare is a
registered trademark of Applied Systems
Technologies, Inc.
© 1989 Applied Systems Technologies, Inc.

Reader Service Number 127

MICRO CORNUCOPIA, #53, May, 1990 13

Figure 4 — DataBear™ Functional Block Diagram c.1990

< SP )

Switched Power

IBM

i-l-^VW^-Xi

4.5V
Maintenance Power

r-*-< <-h —

-===" Power Supply
Switch

8MHz

- f BATT *

100K
100K

*-<

3 x 1.5V
AA

VN2222L

SER.
DATA

1 <— I- NC

2 <-H

.luF

7^ 22p^^

Serial IRQ
Generator

1.5uF

, . 100K

■ j SP ) t-H^w^

10uF

MAX
232 *

cfr

C3+
C1+ C2+
j— *HC1- C2-

3 ^-+

4 <-f

r£

- f 232 *

luFi

4>-

-<■

IBuFr^i 1

C4-

Modul ar Jack

-CUE

10uF

Reset
Generator

n l 10uF

* L

RS232 Interface Chip

232 »

- ( SP

7-40 pF

:r^lh

22p
68HC68

Serial Power Control

- S SP * -

-x

SCK XI
MOSI XO
MISO CE

Mb
Vs

X- LINE

INT
CKO

ex-

Vdd C PUR

POR
Vss PSE

32
kHz

XIRQ\

VCC
EXTAL
*XTAL
3CXIRQ'
i: IRQ'
PRESET/

■ i SUP * ■

- £ PUP * -^*PA5

::PA7H In/Out
28
->'PA6

29,

■ i CSEL * — 33k PA4
31 X PA3.
3S >' PA2 n

33,

4 Sbar * — 34:

- ( RxD * -

< TxD * — ^:

■^c

-24c

24.

25.

R6
100K

* 100K o

Output

Input

*PA1
PA0

* PD0/RxD
PDl'TxD J
PD2/MIS0
PD3/M0SI

* PD4/SCK
^ PD5/SS

SCI

SPI

68HC11

*' MODA

f MODB Wstby
VSS VRH VRL

- C CSEL *

RESETS

Real-Time Clock

1.28V

, k 2.2K

- £ PUP ) e^-vwvH-

100K

LM236
-2.5

Voltage Reference "^

\M^— J 1

14 MICRO CORNUCOPIA, #53, May, 1990

PB7
PB6
PB5
PB4
PB3
PB2
PB1
PB0
PC7
PC6
PC5
PC4
PC3
PC2
PCI

PC0

-A^D

PE7

PE6

PE5

PE4

PE3

PE2

PE1

PE0

,35

,36

,37

,38

■39

.48

,41

d2_

\€>

— «

■48
.46
.44
.49
47

,45_
43

2 MHz

HC00V+-C

EPROM\ = $8000 - $FFFF
RAM\ = $0000 - $7FFF

A0:7

-Q/TT3c

HC373

A8.-12/15

AD0:7

ID
2D
3D
4D
5D
6D
7D
8D

CON

Ql
Q2
Q3
Q4
Q5
Q6
Q7
Q8

EPROMN
27C64

A8:12

A15

I — * . PC

A10

All

A12

PGM\
Vpp

Es^

A8:l2

HC00 P4

< SP * -J

RAMs
6264LP

RAM\

A10

All

A12

Ql
Q2
Q3
Q4
Q5
Q6
Q7
Q8

R/P CE2
CE Vcc

■X-CE3

- J R/UI X

D0:7

iP - CUE

$ Pouer Supply
( BATT * Monitor

Temperature
Sensor

v, F :t

LM34

X PUP 3 — '

SIG.
CONN

+^>GND

SUP - Serial Pouer Control
PUP - Analog Power Control
CSEL - Clock Select

+^>CH 2

n ->swp

^SUP
-§-^>CH 1

Analog Input 2

Pulsed Pur to Ext. Circuitry

Analog Input 1

-f^>GND

Langan Products

2660 California Street

San Francisco, California 941 15

(415)567-8089

MICRO CORNUCOPIA, #53, May, 1990 15

by setting programmable flags referred
to as:

SCAN - Continuous Scan Control,
and;

MULT - Multiple/Single Channel
Control.

A single channel can be sampled once
or continuously; a group of four chan-
nels can be sampled in sequence, either
once or continuously. CCF (conversion
complete flag) indicates there's valid
data.

Because the conversion is quick,
taking just 32 E-clock cycles, I didn't set
up an interrupt for conversion comple-
tion. (Exiting and reentering the inter-
rupt driver would take almost as long as
the conversion, so why bother?)

Any unused A/D channels can
double as high-impedance digital inputs
(Port E).

Real Signals

I've installed an LM34 (or LM35)
temperature sensor on the printed cir-
cuit board. The LM34 is a linear, positive
temperature coefficient sensor. It out-
puts a voltage that's proportional to
temperature from -50 to +300 degrees F.
With the LM34, the output is lOmV/de-
gree F with 1.000 Volts at 100 degrees F.
Easy, huh?

Supply anywhere from 4 to 30 Volts
to the sensor and connect it to the 6811
A/D Converter and you've got tempera-
ture!

Although there are more and more
linear sensors with the conditioning elec-
tronics built-in, most aren't as easy to
use. Many outputs range between and
0.1/0.5/5.0/10.0 Volts, to 1 mA, 4 to 20
mA, strain gage mV/V-type outputs,
frequency, pulse-per-event, etc. You'll
have fun monitoring a collection of these
outputs with a single system.

Mainframe loggers, such as those
from Fluke and Hewlett-Packard, have
module connectors for signal condition-
ing assemblies. These assemblies are
quite expensive, and designing a logger
to handle all the standard modules can
make the logger expensive as well.

I wanted the DataBear to log any
variable, any signal, as long as it was be-
tween and 1 Volt! Obviously I needed
to add some signal conditioning.

For example, there's a standard rela-
tive humidity sensor package for the
Bear. The sensor is a half-bridge humid-
ity-to-strain transducer. You apply
power and it spits out a humidity-pro-
portional voltage.

Figure 5 — Interrupt Vector Assignments

Vector Address

Interrupt Source

Local Mask

FFCO, CI

FFD4,D5
FFD6,D7

Reserved

SCI Serial System

I Bit

See Table A
Below

FFD8, D9
FFDA,DB
FFDQDD
FFDE,DF

SPI Serial Transfer Complete
Pulse Accumulator Input Edge
Pulse Accumulator Overflow
Timer Overflow

I Bit
I Bit
I Bit
I Bit

SPIE
PAII
PAOVI
TOI

FFEO, El
FFE2,E3
FFE4, E5
FFE6, E7

Timer Output Compare 5
Timer Output Compare 4
Timer Output Compare 3
Timer Output Compare 2

I Bit
I Bit
I Bit
I Bit

OC5I
OC4I
OC3I
OC2I

FFE8,E9
FFEA, EB
FFEC,ED
FFEE, EF

Timer Output Compare 1
Timer Input Capture 3
Timer Input Capture 2
Timer Input Capture 1

I Bit
I Bit
I Bit
I Bit

OC1I
OC3I
OC2I
OC1I

FFF0,F1
FFF2, F3
FFF4, F5
FFF6, F7

Real Time Interrupt

IRQ (External Pin or Parallel I/O)

XIRQ Pin (Pseudo Non-Maskable Interrupt)

SWI

I Bit
I Bit
XBit
None

RTII

See Table B

None

FFF8, F9
FFFA,FB
FFFCFD
FFFE,FF

Illegal Opcode Trap
COP Failure (Reset)
COP Clock Monitor Fail (Reset)

None
None
None
None

None
NOCOP
CME
None

RESET

Table A
SCI Serial System Interrupts

Interrupt Cause

Local Mask

Receive Data Register Full

RIE

Receiver Overrun

RIE

Idle Line Detect

ILIE

Transmit Data Register Empty

TIE

Transmit Complete

TCIE

Table B
IRQ Vector Interrupts

Interrupt Cause

Local Mask

External Pin

None

Parallel I/O Handshake

STAI

16 MICRO CORNUCOPIA, #53, May, 1990

So I built an op-amp circuit which
would power the sensor, amplify its out-
put, and remove the resulting offset so
that the final signal lay between and 1
Volts. The Bear controls power to the in-
terface's electronics; it's turned on before
each sample is taken and removed after-
wards.

Each type of sensor required its own
interface electronics. There are many
good books available on interfacing sen-
sors to A/D Converters. (Analog Dev-
ices has a good series of texts available.
National and Motorola both have loads
of good data sheets and application
notes for little or no charge. The Op-Amp
Cookbook by Walter Jung is also full of in-
terface ideas.)

The 6811 makes it easy for the de-
signer by accepting different high and
low voltage references, extremely high
input impedance, built-in sample-and-
hold circuitry, and an eight-channel
multiplexer. All are accessible and con-
trollable via CPU registers.

Parallel I/O

Depending on the flavor of 6811, you
have as many as 38 parallel I/O lines.
However, in the expanded, multiplexed
address bus mode, with ports B and C
working as the address /data bus, you
cut I/O to 22 lines.

The SCI port uses 1 or 2 lines and the
SPI port uses 4. Because the DataBear
uses both the SCI and the SPI, we're left
with 16 lines. I reserved the A/D Con-
verter channel inputs for analog, leaving
just 8 parallel I/O lines.

Port A's I/O is closely linked to the
advanced timer functions built into the
6811 (see the section on Timers). Some of
the Port A lines are dedicated inputs
(PAO, 1, and 2), others are dedicated out-
puts (PA3, 4, 5, 6), and the remaining pin
(PA7) is bidirectional under program
control.

In the single-chip mode, ports B and
C become available for parallel I/O. Port
B is all output. Port C is bidirectional.

Port D (which has only six lines) is
split between bidirectional I/O, SCI port
(RD and TD), and the SPI Port (SCLK,
MOSI, MISO, SS). All, or part of, the I/O
lines can be used as parallel I/O de-
pending only on your uses of the SCI
and SPI systems.

I/O pins can sink from 1 to 10 mA,
depending on how much current adja-
cent pins sink, but they source only
about 1 mA, max. If you want to control
any significant amount of current, you'll

need to add a buffer chip or transistor. I
have used the 6811 to drive a sensitive
relay, but it's not a good idea.

I used a parallel I/O line to control
the current to the MAX232 serial chip.
The chip can pull around 30 mA when
the charge pump starts up. (It's always
extra work to prime a pump.) The Serial
Power Control section in Figure 4 shows
how a ten cent 2N3904 transistor han-
dles the MAX232's current flow. The
ULN2001 series driver chips also do a
good job of driving current-gobblers.

Timers

One of the nicest features of the 6811
is its array of timers and counters. The
general architecture is based on the
MC6840 counter-timer, with improved
and enhanced control and interrupt
capabilities. The 6811 can act as a
frequency counter, a waveform gener-
ator, or period timer. If you're clever you
can even make it do combinations of
these if you pay careful attention to sub-
tleties like interrupt latency and counter
rollover.

The Pulse Accumulator Input (PAD
operates through pin 7 of Port A. Transi-
tions on pin 7 clock an 8-bit counter.
Counter rollover can generate an inter-
rupt which you can also count, thus ex-
tending the counter to 16-bits or beyond.
I've used the PAI function up to 50 KHz.
It's rated to count at one-half the E-clock
frequency.

The Input Capture Functions (ICF) let
you time events (transitions) via a 16-bit
counter running at the E-clock
frequency.

Output Compare Functions (OCF) do
the opposite — they cause a level transi-
tion at an external pin whenever their
control register matches the value of the
16-bit free-running counter.

All the features are linked to the in-
terrupt system, with reassignable priori-
ties. I can hear the wheels turning out
there, especially among the servo-motor
guys! While the DataBear doesn't make
much use of the timer functions, they are
a 6811 crown jewel.

Low Power Modes

Most CMOS processors will shift to a
low-power mode or modes under pro-
gram control. The 6811 has two low-
power modes: WAIT and STOP.

In WAIT, the E clock oscillator con-
tinues for the duration of the WAIT. Plus
you can choose which subsystems stay
active.

During STOP the E clock oscillator
stops, shutting down all internal sys-
tems, such as the SCI and SPI.

Current consumption in WAIT can be
as low as 6 mA, depending on the pro-
cessor modes and number of peripherals
still running. In STOP mode the current
consumption drops to well under 1 mA.

It takes an interrupt or a RESET to
wake the processor from either low-
power mode.

In WAIT mode, the interrupt can be
an incoming serial data character or a
timer timeout. The SCI subsystem also
has a special wake-up feature which al-
lows an active serial RD line to wake up
the processor.

When the processor's in STOP mode,
the interrupt must come in from outside.
The DataBear's processor spends ap-
proximately 99.5% of its time in STOP
mode, so I've included three sources of
external interrupts: the clock chip
(XIRQ), the power source switch (IRQ),
and a FET connected to the RD serial
line (also IRQ).

Clock interrupts normally occur once
a second and take priority over other in-
terrupts. I use the clock interrupt as the
Bear's time base, driving the data sam-
pling. In Continuous Mode, the clock in-
terrupt is always every second so
sample rates are multiples of one sec-
ond. In Fixed Mode, the clock interrupt
occurs at intervals as short as Vi28 th of a
second during the programmed sam-
pling bursts.

Serial interrupts will wake the Bear
from dead STOP by pulling the IRQ line
low. +12 Volts on the RD line turns on
the VN2222 FET, which is wire-ORed
with the open-drain status output of the
ICL7663 power source switch.

The ICL7663 power source switch
also pulls the IRQ line low whenever the
supply goes higher than the main-
tenance level. The DataBear is able to de-
termine which interrupt awakened it by
the context in which the interrupt oc-
curred and by some status checks in the
interrupt handler.

Power

For battery-powered systems, like
this, a CMOS microprocessor offers two
major advantages: low power consump-
tion, and the ability to operate from a
range of voltages. The 6811 consumes
approximately 20 mA operating on 3 to
7 Volts.

Using batteries sounds like a piece of
cake until you consider the more subtle

MICRO CORNUCOPIA, #53, May, 1990 17

issues. First, batteries develop internal
resistance as they discharge so the volt-
age drops. Although open-circuit volt-
age may be fine, it may drop below
minimums as the processor turns on.

I gave the Bear a 4.5 Volt main-
tenance supply to allow an orderly ces-
sation of sampling as the output of the
regular battery drops. An Intersil
ICL7663 selects whichever battery has
the higher output.

The second and third complicating
factors are switchover hysteresis and the
level at which regular operation will
cease completely. The DataBear pre-
serves RAM data as the batteries run
down by ceasing regular operation. The
clock interrupts are turned off and the
time of power loss is stored. With any
luck, the user will retrieve the Bear
before the maintenance supply dies.

You'll find there are many types of
batteries: zinc-air, lithium, mercury,
silver-oxide, ni-cad, and alkaline. In fact,
there are so many options that batteries
would make an article in themselves.

I finally chose alkaline AA cells be-
cause they're cheap, easy to find, and
pack a fair number of Amp-hours.

Battery supplies have been the most
difficult part of the system. I've planned
to use the Bear for long-term, remote or
unattended field measurements, which

68000

SK*DOS - A 68000/68020 DOS containing
everything you expect in a DOS - on-line help,
multiple directories, floppy and hard disk sup-
port, RAM disk and/or disk cache, I/O
redirection, and more. Supplied with editor,
assembler, Basic, powerful utilities. Supported
by Users' Group and BBS. Software available
from other vendors includes C compiler,
Basic, editors, disassemblers, cross-assem-
blers, text formatter, communications
programs, etc. Priced at $165 with configura-
tion kit, less if already configured for your

system.

HARDWARE - 68xxx systems

start at $200. Call or write

7
Star-K Software
Systems Corp.

P.O.Box 209

Mt.Kisco NY 10549

(914) 241-0287 / Fax (914) 241-8607

makes the power supply issue very sig-
nificant. My best suggestion is that you
obtain as much technical information on
the batteries you plan to use in your sys-
tem, then make your selection.

Evaluation Module

Developing a PROM-able program
for any microprocessor can be difficult,
no matter which processor and which
development tools. Unless you use com-
mercial hardware, even the most basic
system resources must be tested and de-
bugged before you can start writing soft-
ware.

Fortunately, Motorola provides low-
cost development hardware for their mi-
croprocessors. There are two systems
available for the 6811 family: the
6811EVM, and the 6811EVB.

The -EVB costs less than $100 and
comes with the BUFFALO monitor in
ROM. It also comes with the 68HC24 so
you can use all 38 parallel I/O lines, as
well as the external PROM and RAM.
The memory space available is 64K, with
the monitor PROM in the upper 8K at
$E000. The -EVB's manual includes code
examples.

The -EVM will run you between $400
and $500 but has more complete debug-
ging support. They've thrown in a moni-
tor with this board, too.

I used a 6811 EVB to do the initial
development for the DataBear. That way
I could concentrate on getting the soft-
ware running without worrying about
the hardware.

I had to learn how to use the -EVM,
but that's a lot easier than learning
developing from scratch. I built all the
custom hardware on a small wirewrap
board and connected it to the expansion
connector on the -EVB.

By downloading it to the -EVB (via
the monitor) and carefully setting break-
points, I was able to debug the code
without emulators. I used the Ar-
chimedes cross-assembler running on
my DOS system.

The initial prototype consisted of the
custom wirewrap board and the -EVB
sandwiched together and squashed into
a plastic box. Though the combination
was about four times the size of the final
Bear, clients and customers were en-
thusiastic!

My message is, it can be done on a
nonexistent budget with simple tools.

Vectors

Once you start developing code for a

standalone 6811, you have to deal with
certain facts. First, define the event vec-
tors. The event vectors are addresses of
code routines which the processor ex-
ecutes in response to certain events, such
as interrupts or resets.

The 6811 supports all the usual vec-
tors, such as the external hardware inter-
rupts and reset. It adds vectors for all the
timer functions, SCI and SPI ports, real-
time and watchdog timers, illegal in-
struction, and software interrupts (SWI).
For the 6811, the vector table lives at
$FFD0 through $FFFF. The 6811 reset
vector lives at $FFFE:FF.

You'd better have something in mind
for every single vector, even if it's only
an RTI (Return from Interrupt). Proto-
types often suffer from spurious inter-
rupts which can be a bear (no pun
intended) to isolate.

Relocatable Registers And RAM

The 6811 must be initialized within
64 E-clock cycles of a reset. The most im-
portant is mapping the processor inter-
nal registers and internal RAM to the
desired 4K ($1000) page boundary.
Write the locations of the RAM and reg-
isters to the INIT register in the first set
of instructions following reset.

You can use the init code to deter-
mine: whether the IRQ interrupt is level
or edge sensitive, the oscillator delays
after STOP, and watchdog timer periods.
Take a special look at the INIT, CON-
FIG, and OPTION register descriptions
in the Motorola literature.

EEPROM

Some versions of the 6811 include 512
bytes (or more) of EEPROM. This can be
particularly useful for storing serial
numbers, network or board addresses
and IDs, configuration or password
data, etc. The EEPROM is located at a
fixed memory address. If you have EE-
PROM, make sure you can instruct your
linker to avoid its location.

Miscellaneous Notes

I won't try to cover everything, but I
should mention a few tricks I discovered
during this project.

First, I used a current probe and an
oscilloscope to find out which subsys-
tems were drawing significant power
when they weren't being used. Helped
cut power use.

Second, I found it's important to let
external devices stabilize (after the pro-
cessor comes out of STOP) before sam-

Reader Service Number 40
18 MICRO CORNUCOPIA, #53, May, 1990

DDBD D

BUB B n

BBBB B BBBUBB

BBBBBB BBDBBBBBB

BBBBOB BBBBBBBBOaa

nnmaBBBBHiiaHDiiaanB

BBBBBBBBBBBBBBBOBBB
BBBBBBBBBBBBBBBUBBB
BBB BBBBBOBBBBBBBBD
BBBBBBBBBBBBBBBBBB
BBBBBBBBBBD BBBBBB
BBBBBBBBB BBBBBB
BBBBBBBBB BBBBBB
BBBBBBBO BBBBBB BB
BBBBBBBB BBBBB BBBB
BBBBBBBD BBBSBBBBBBBBB
BBBBBBBB BBBBBBBBBBBBBBB
BBBBBBB BBBBBBBBBBBBBBB
1UBBBBB BBBBBBBBBaflflBBBBBB
IBBBBB BBBOBBBBBBBBBBBBBBBB
BBBBBBBflBBBBBBBBBBBflflfl
BBBB BBBBBBBBBBBBBBBBBB
BBBBBBBBBBBBBBBBBBBBBB
BBBBBBBBBBBB BBBBBBBB
BBBBBBBBBBBBB BBBBBBB

BBBBBBBBBB BBBBBBBB
BBBBBBBBBB BBBBBB

BBBBBBBBB BBBBBB
BBBBBBBBB BBBBBB
BBBBBBBB BBBBB
BBBBBBB BBBBBB
BB BB BBBBB
BBB BBBB
BB BBBB
BBBB
BBBB

wc,

Add lightning fast graphics to your programs quickly and easily through the popular
PCX file format. Why reinvent the wheel? Make your programs immediately compatible
with hundreds of packages from Aldus PageMaker to ZSoft's PC Paintbrush with these
linkable graphic libraries.

"An exceptional product" - Programmer's Journal, Aug '89 $195

Version 3.5 of the PCX Programmer's Toolkit gives you over 60 powerful functions to
manipulate bitmapped graphics. Use Virtual screens, Super VGA modes, LIM 4.0 support,
a 300 page manual, 9 utilities including screen capture and display, and the fastest routines
on the market.

Need Special Effects, but caught in a GRASP? $99

Why create a demo when you can create the real thing? Don't be trapped in a slideshow
editor or demo program when you can use PCX Effects for the PCX Toolkit and your favorite
programming language. A Music Language and spectacular effects for exploding your
graphics!

Blazing Graphics Text $149

With PCX Text you can display text with graphics as fast as it always should have been.
Display characters, strings, fixed and proportional text, background transparency, and
more. Includes a font editor, 85 fonts, and text utilities for blazing graphics bitmapped text.

NO ROYALTIES!

30-DAY MONEY-BACK GUARANTEE!

SOURCE CODE AVAILABLE!

All packages support 12 compilers for C, Pascal, Basic, Fortran,
Assembly, and Clipper. All modes of the Hercules, CGA, EGA, VGA,
and Super VGA adapters are supported, up through 800x600x256
(22 modes in all). Assembly Language source code is optionally
available. Trademarks are property of their respective holders.

MICROPROGRAMMING

11315 Meadow Lake • Houston. Texas 77077 • (713) 870-0737

FAX: 71 3-870-0288
Add $5/package for UPS Ground
Texas Residents add 8% Sales Tax

_.. . . i

For orders and information, call:

■227-

VISA/MC/AMEX/COD/CHECK/PO

J ,™™.™^« ,

WjMU^rM.VMWH^NWUHWn.uWM.^WKW.Mv.MUmn^niHU^

pling data or sending them instructions.
This is important because low-power
analog systems can take several hundred
microseconds to wake up.

Third, I found that the input im-
pedance of the 681 l's A/D Converter is
so high that open, unused inputs will
show significant voltage. Tie unused in-
puts to ground through a 100K resistor.

Fourth, with any low-power system
you'll need to pay attention to input im-
pedance. Pullups should have a pretty
high resistance; 100K works well.

To Boldly Go...

So what's the Bear doing?

It's logging blast furnace temperature
data for specialty melts, monitoring
freezer temperatures, and recording
weather data.

It has been stuffed inside a fresh fish
during air shipment ("Bear Eaten by
Fish!").

It's been taken to the opera (you get
drowsy in the balcony when it's 80+
degrees up there), and hung on grape
vines in Napa Valley. It's even measured
actual earthquake weather in San Fran-
cisco!

The Bear Of The Future

Where does it go from here? For start-
ers, like any creative designer, I want to
take advantage of more features of the
6811. I'd like to use the frequency count-
ing and pulse counting features of the
PAI system so the Bear could log digital
data. It would be nice to have the Bear
log the time of events, not just at regular
intervals. And, of course, there are thou-
sands of sensors that can tie right into
our to 1 Volt standard input.

Eventually I expect to enter the world
of surface-mount, volume production. I
will have to do a new circuit board and
find a whole new list of suppliers and
packages to make the Bears.

If I Knew Then What I Know Now

No design article would be truly
complete without a True Confessions
section. Hindsight being what it is, I cer-
tainly would not have done the whole
package in assembly. As I write this,
other programmers are adding software
features. It would be much easier now if
I'd developed in C from the beginning.

I made it much harder on myself in
the battery department by not seeking

expert advice at the beginning. We've
learned some things about batteries that
cost time and money. (My money.)

I am thankful that I used a processor
which I knew well. There are many sub-
tleties in the use of microcontrollers.
They are powerful critters, but you can
wind up with powerful headaches if you
ignore their little gotchas. Take the time
to learn their ways and you can do a lot.

References

Motorola 68HC11A8 HCMOS Single-
Chip Microcontroller Manual, Motorola
Document Number MC68HC11A8/D.

Motorola 68HC11A8 HCMOS Single-
Chip Microcontroller Programmer's Refer-
ence Manual, Motorola Document
Number MC68HC11PM/ AD.

♦ ♦ ♦

MICRO CORNUCOPIA, #53, May, 1990 19

When you're up to your knees in a new
issue (this one), the last thing you want to
deal with is a bunch of snakes. Fortunately,
these snakes won't object to GRAS (that's
Generally Regarded As Safe). Now, whose
system would benefit from a case of slithers?

Understand now, I set out to do
what I was supposed to — build a
voice-control system using the
Motorola VCP-200 (Radio Shack #276-
1308). This chip will recognize five
phrases independent of the speaker: go,
stop, turn left, turn right, and reverse. Its
data sheet (from Radio Shack) includes a
schematic for the chip's circuit. All I
needed to do was connect the chip to the
microphone and analog amplifier /filter
from Micro C #50.

But I didn't do it yet. I became dis-
tracted trying to imagine how I'd use my
apparatus. (I should have listened to
Larry, who never gets bogged down in
these kinds of details.)

So I digressed — what was I to do
with the signals from the chip? How
could I use commands like: "left turn"
and "reverse?"

I speculated — draw something on a
screen: an entity which moved under
voice control (leading perhaps to a vocal
PAC MAN). But what entity?

Then (as though in a vision on a hot,
hazy day in the high desert) it came to
me: snakes. A snake (an object) is a
bunch of linked segments (also objects).
To move the snake forward, I'd create a
new segment, call it the head, then tell it
(the lead segment) to send a message
back to the tail saying "add a new seg-
ment" or "drop a segment."

So I built snakes instead of connect-
ing hardware. In the process, I generated

some interesting C++ code and utilized
several C++ 2.0 features. SNAKES, my
code creation, is (of course) a riot.

When you run SNAKES, tell it how
many snakes you want, how long you
want each snake to be, and how quickly
you want to see action. By increasing the
delay factor, you can slow things down
as much as you like. Each has its own
ASCII character.

I compiled SNAKES using Zortech
C++ version 2.06 and their disp package
to increase display speed. In the Micro C
listing, I've #ifdefed disp out so SNAKES
will work (albeit slower) with other im-
plementations of C++. (These include
UNIX as long as you have an ANSI ter-
minal, or are willing to rewrite the brief
segment::draw() and segment::erase()
routines.) You can also easily change
these routines to support graphics.

Classes

Figure 1 (SNAKE.HPP), the header
file, contains the class definitions for
point (which holds x-y information),
segment, and snake. I made point a
struct (a class with all members public)
because I wanted the convenience of
having coordinate information packaged
together without having to bother with
private data.

The constructor simply initializes the
data elements x and y using the initial-
izer list. (That is, you can treat built-in
types as if they had constructors in the
initializer list only.)

We overload the == operator to deter-
mine whether one point is equivalent to
another. This prevents a snake from
backing over itself.

random_point() is a static member
function which returns a random point.
random_point() creates a temporary ob-
ject by calling the point constructor.

Note: we don't need an identifier since
the object (a point) immediately returns.
A static member function addresses
the class as a whole, not a specific in-
stance of the class. This could have been
an independent (non-object contained)
function, but I made it static so anyone
using class point can now see (and
access) it. Also the identifier ran-
dom_point is local to class point, and
doesn't "pollute the global name space."
For this and other reasons, static mem-
ber functions are an important addition
to C++ 2.0.

Segments

The segment class contains a pointer
to the previous segment in the list. So
any segment can send a message to the
segment behind it ("drop the tail seg-
ment, add a segment").

The point object lets the segment lo-
cate itself on the screen. It draws itself
and prevents its own redundancy — the
entire snake slithering over itself (al-
though other snakes may slither over it).

segment displays itself on the screen
using a pattern and orients itself through
a variable called heading of the
enumerated type direction, heading de-
termines a snake's direction and tries to
keep it going that way.

(Aside: my original design didn't include
heading. I added it after I got the system
working and saw snakes doubling back on
themselves, creating an ugly mess. For-
tunately, it was a simple matter to add head-
ing to the existing system. In general, try to
get the system up quickly. You can't know
some things while you're working on the de-
sign, only after you see the design working.

Local Enumerations & Anti-pollution

Notice that enum direction is inside
the class definition. In C++ 2.0, enumera-

20 MICRO CORNUCOPIA, #53, May, 1990

By Bruce Eckel

Revolution2

P.O. Box 760

Kennett Square, PA 19348

Figure 1 — SNAKE.HPP

#include <stddef.h> // size_t definition

struct point { // coordinate representation
int x, y;

point (int xi = 0, int yi = 0) : x(xi), y(yi) {}
int ope rat or== (point rval) { // test equivalence

return x = rval.x && y == rval.y;
}
static point random_point ( ) ; // generate random point

};

class segment {

segment * previous;

point sp; // position on screen

char pattern; //to display on the screen

enum direction heading; // "tendency of motion" of the snake

void erase () ;

void checkheapO; // make sure this is on the heap!
public :

// this enumeration is local to the class:

enum direction { UP, DOWN, RIGHT, LEFT };

void draw() ;

// re-define operator new for this class ONLY:

void * operator new(size_t sz) ;

segment (point p, char ptrn);

segment (segment * prv, direction dir) ;

~segment() { erase (); }

int shed_tail(); // go back to the tail and drop it off

point seg_point() { return sp; }

int cross_over (point) ; // is this point an existing segment?

direction path() { return heading; }

void redraw_tail() ; // redraw everything from here back

};

class snake {

segment * head;

int length, maxlength;
public:

snake (point p, int size, char ptrn);

void crawl (direction) ;

void slither ();

};

hen you run
SNAKES, tell it how
many snakes you
want, how long you
want each snake to
be, and how quickly
you want to see
action.

tions are local to classes. The enumera-
tion tag is global (everyone can refer to
direction). But the enumeration names
are local to the class, so they don't pol-
lute the global namespace — you must
refer to them as segment: :UP, etc.

Preventing namespace pollution is a
very important feature of object-oriented
languages. Big projects often become un-
manageable simply because they create
more names than they can keep up with.

You can also use enumerations local
to a class to create "local const" values.
You can define a const inside a class, but
you can't initialize it when you define it
(unlike its definition everywhere else,
when you must initialize it). Do it in-
stead in the constructor initializer list
(thus you can end up with different
values for the const depending on which
constructor you call).

However, you can create the equiv-
alent of a local const and initialize it at
the same time using an untagged enum,
like this:

class x {
MICRO CORNUCOPIA, #53, May, 1990 21

enum {red=7, blue=4}; //no tag!
// ...

};

Now you can use the identifiers red
and blue just like const values. Well, al-
most — at the time I wrote this, different
compilers exhibited different behaviors
when I tried to use red as a dimension in
an array definition. I'm not sure how it's
supposed to work. Note that you can
never use even a const as an array di-
mension in ANSI C.

Class Overloading Of "new"

Another feature of C++ 2.0 lets you
overload the operators new and delete
(which control dynamic memory alloca-
tion) on a class-by-class basis. Normally
you do this when you want to create dy-
namic objects more efficiently. I've done
it here because segment objects must
never be created on the stack, only on
the heap.

To signal the programmer, seg-
ment-operator new() (shown in Figure
2) puts a "signature" in the dynami-
cally-allocated memory. The function
segment::checkheap() looks for this sig-
nature. If it doesn't find the signature, it
sends a message to the programmer. It
isn't foolproof, but will usually detect
misuse. All segments must be created on
the stack because they're only destroyed
by segment: :shed_tail() when it calls de-
lete.

"segment" Constructors

segment has two constructors. The
first — for the initial segment of the
snake — establishes the starting point and
the pattern to be used. The second
makes a segment which is linked to an
existing segment, in a specified direc-
tion. Note that both constructors call
checkheapO to insure the object was
created on the heap.

The first constructor (Figure 2) is
straightforward, but the second must
"wrap" the coordinate of the new seg-
ment around the display screen if it goes
off the edge. Also, the draw() method
isn't called in the second constructor,
since a new snake may need a new seg-
ment which crosses over an old segment
of itself. The calling function must check
the new segment before actually draw-
ing it.

Recursive Member Functions

The next three functions are pure fun.
They recurse back down the snake

Figure 2 — SNAKE.CPP

# include "snake . hpp "
#include <stdlib.h>
# include <stdio.h>
# include <time.h>
#include <string.h>

// assuming ANSI terminal or ANSI. SYS on a PC

// memset()

#ifdef ZTC // use disp package w/ Zortech

# include <disp.h>

# include <conio.h>

#endif

const height = 23; // screen height

const width = 78; // screen width

point point : : random_point ( ) {

return point (rand () / (RAND_MAX/width) ,

rand() / (RAND_MAX/height) ) ;
}

void * segment :: operator new(size_t sz) {

void * tmp = : :new unsigned charfsz]; // allocate with global new()
memset(tmp, ' x' , sizeof (segment) ) ; // fill with x's
return tmp;

}

void segment: : checkheapO {

if (pattern != 'x') { // check for memset pattern

fputs("can only create segments on the heap!", stderr) ;
exit(l) ;
}
}

segment : : segment (point p, char ptrn)

// capriciously choose old heading:

: sp(p), previous (NULL) , heading (LEFT) {

checkheapO ;

pattern = ptrn;

draw ( ) ;
}

segment: : segment (segment * prv, direction dir)

: previous (prv) , sp (prv->sp) , heading (dir) {

checkheap () ;

pattern = previous->pattern;
// if(pattern++ = ' z' +1) pattern = 'a'; // pattern test...

// create new segment in appropriate direction & wrap:

switch (dir) {

case

UP :

if (~sp.y

0) sp.y = height;

break;

case

DOWN :

if (++sp.y

height) sp.y = 0;

break ;

case

LEFT :

if ( — sp.x

0) sp.x = width;

break;

case

RIGHT :

if (++sp.x

width) sp.x = 0;

break ;

}

// Recursive function to go back to the beginning of the
// sequence of segments, to find the tail and remove it:
int segment : :shed_tail () {
if (previous = NULL) {

delete this; // NULL means we're at the tail

return 1; // indicates to the next call 'up' that the tail was
found
}

// recursive call until tail is found:
if (previous->shed_tail() ) // 1 means the call removed the tail

previous = NULL;
return 0; // means the call didn't remove the tail
}

// Recursive check to see if p crosses an existing segment:
int segment : : cross_over (point p) {

if (previous == NULL) return p == sp;

if (p == sp) return 1;

// recursive call to go to the end, or a cross point:

return (previous ->cross_over (p) ) ;
}

// Recursively redraw entire snake:
void segment : :redraw_tail () {

draw() ;

if (previous == NULL) return;

Continued on page 23

22 MICRO CORNUCOPIA, #53, May, 1990

Continued from page 22

previous->redraw_tail () ;

}

{

void segment : : draw ( )
#ifdef ZTC

disp_move (sp . y, sp . x) ;

dispjputc (pattern) ;
#else

printf ("\xlb[%d;%dH", sp.y, sp.x);

put char (pattern) ;
#endif
}

void segment :: erase () {
#ifdef ZTC

disp_move (sp.y, sp.x);

disp_putc (' ' ) ;
#else

printf ("\xlb[%d;%dH",

put char (' ' ) ;
#endif
}

// move cursor

sp.y, sp.x); // move cursor

snake: : snake (point p, int size, char ptrn)

: length (0) , maxlength(size) {

head = new segment (p, ptrn) ;
>

// Just mindlessly goes where you tell it:
void snake : : crawl (direction dir) {
head = new segment (head, dir) ;
if (++length >= maxlength) {
--length;

head->shed_tail () ;
}
}

// find it's own way, tending to the current path:
void snake: : slither () {

const int chance = 10; // 1 in 10 chance to change directions

direction old = head->path () ;

// Choose a new direction by first throwing the dice to decide

// whether to actually change direction. If so, choose a new

// direction randomly:

if (rand()/(RAND_MAX/ chance) != chance/2)

direction dir = old; // change only if magic number is rolled
else {

dir = (direction) (rand() / (RAND_MAX/4) ) ;

head->redraw_tail () ; // occasionally redraw everything
}

segment * new_seg = new segment (head, dir) ;

// check that the new segment doesn't cross over the existing snake:
if (head->cross_over (new_seg->seg_point () ) ) {

delete new_seg;

return; // new direction causes cross-over
}

// successfully found new direction
head = new_seg;
head->draw() ;
if (++length >= maxlength) {

— length;

head->shed_tail () ;
}

static void pause (int rate) {
for (int i = 0; i < rate; i++)

}

main (int argc, char * argv[]) {
if ( argc < 4 ) {

f puts ("usage: snake num_of_snakes size_f actor pause_rate\n"

"CTRL-BREAK to quit\n", stderr) ;
exit(l) ;
}

printf ("\xlb [=71") ; // turn off ANSI line wrap
#ifdef ZTC

Continued on page 24

towards the tail. It's as if you're tra-
versing a linked list, with each link an
object with its own member functions.
This is inspired by the much more
sophisticated "agent" example in Ravi
Sethi's Programming Languages — Concepts
& Constructs (Addison-Wesley, 1989). If
the rest of the book is anything like this
example, it's a gem.

The function segment: :shed_tail()
must do three things. If the current seg-
ment is the tail (in which case the pre-
vious pointer will be NULL), then it
deletes the tail (thus the reason for forc-
ing segments to be made on the heap).
However, if the tail was just deleted,
then the segment just ahead of the tail is
the new tail.

Thus, when the tail is deleted, the re-
cursive function returns a 1 to tell the
next segment up that it should set its
previous pointer to NULL to indicate it's
the new tail. Finally, if nothing happens,
it returns a to tell the next segment to
do nothing but return in turn.

Don't Tread On Me!

segment::cross_over() takes its argu-

Unique

Programmer's
Books & Tools for
MS-DOS & OS/2

Professional function
libraries and advanced books
by Michael J. Young. Create
high-performance MS-DOS

programs, TSRs, OS/2

programs, and Presentation

Manager applications.

Items not available

anywhere else!

C and assembly language!

Please call or write for
free catalog:

Young Software Engineering

20 Sunnyside Avenue, Suite A

Department MC01

Mill Valley, CA 94941

415/383-5354

Reader Service Number 207

MICRO CORNUCOPIA, #53, May, 1990 23

ment, a point, and checks to see if it's
equivalent to the segment's point. If so,
it returns 1 to indicate yes (which means
the point isn't a good choice). If not it
calls previous->cross_over() to check
back until it finds a crossover point or
the tail.

segment::redraw_tail() calls drawO,
then recursively calls itself until it en-
counters its tail, thus redrawing the en-
tire snake. draw() and eraseO are the
only functions that output anything. So
if you want to adapt the program to a
new system, you only have to change
these (and a couple of minor places in
mainO).

The "snake" Class

The snake (Figure 1) is quite simple.
A snake consists of a pointer to a seg-
ment called head, and numbers indicat-
ing the current length and maximum
length. A single constructor (Figure 2)
creates a new segment for the head, and
initializes length and maxlength. After
that, the snake will crawlO in any direc-
tion you say (for the voice command
system).

snake::crawl() simply assigns head to
a new segment made from the previous
head (thus chaining the snake together).
If the snake is too long, it sheds its tail.

Chance Slithering

snake::slither() is a more interesting
function. It "throws dice" using the
ANSI C library function rand(), which
generates random numbers, to decide
whether to change directions. Note that
RANDJVIAX is a constant defined in
stdlib.h which is the maximum value
randO will return. Thus the statement:

if (rand() / (RAND_MAX/chance) •=
chance/2)

creates a random number between and
chance, and tests to see if it's equivalent
to chance/2. Thus, in a perfect distribu-
tion you have a one in chance prob-
ability the statement will be true.

Notice that normally you would
think like this: rand()/RAND_MAX re-
turns a number between and 1, so just
multiply that by chance to get a number
between and chance. But this would be
wrong. Since the compiler will do in-
teger division, rand()/RAND_MAX will
almost always return 0!

Thus you must force RAND_MAX to
be divided by chance first to insure in-
teger division doesn't mess up the an-

Continued from page 23

disp_open() ;

disp_move (0,0);

disp_eeop() ;
#else // ANSI terminal

printf ("\xlb[2J") ; // clear screen
#endi£

// seed the random number generator:

time_t tnow;

time (&t now) ;

srand(tnow) ;

// Create a whole nest of snakes:

const num_of_snakes = atoi (argvfl] ) ;

const size_factor = atoi(argv[2] ) ;

const pause_rate = atoi (argv[3] ) ;

snake ** snakes = new snake* [num_of_snakes] ;

for(int i = 0; i < num_of_snakes; i++)

snakes [i] = new snake (point : : random_point () ,
#if // for randomly- chose sizes:

rand() / (RAND_MAX/size_f actor) , i + '0');
#else // for fixed sizes: (avoid the bell)

size_factor, ((i + 1) % 254 == 7)
#endif
#ifdef ZTC

while (!kbhit()) {
#else

while (1){
#endif

for(i = 0; i < nura_of _snakes ;
snakes [i] ->slither() ;
pause (pause_rate) ;

(i + 1) % 254),

i++) {

}

Figure 3 — PAUSE.C : Portable function for delays

#include <time.h>

/* This is a pure ANSI C function, so it's portable: */

void pause (int hundredths) {

clock_t start_time = clock (); /* current time */
while (( (clock () - start_time) * 100) /CLOCKSJ?ER_SEC <
; // wait in this loop for the delay time

}

hundredths)

swer. This logic works in any situation
(there are others in SNAKE) where you
generate a random number in a particu-
lar range.

If the above "if" statement is true,
snake::slither() will slither the snake in
the same direction by using the old
direction. (Thus you can make the
snakes do more twists and turns by re-
ducing chance.) If it's false, generate a
random number between and 4 and
cast it into a direction (to indicate one of
the directions in segment).

Note that C++ 2.0 forces you to cast
an integer into an enum explicitly; it
won't just allow you to sneak it by — bet-
ter type checking!

Next, we create a new segment, but

we don't assign it to the head — yet. First,
we test to insure that the new segment
won't cross over an existing segment. If
it does, we delete the segment, and the
function returns without doing any-
thing. The next call to slitherO will try
again with a new direction.

When running the simulation, you'll
see that sometimes a snake will coil in
upon itself and be unable to move. (If
you run the simulation long enough, all
snakes will eventually end up this way.)

A Flexible main()

In The Tao of Objects (a book I am
coauthoring with Gary Entsminger), and
we maintain that one of the fundamental
tenets of object-oriented programming is

24 MICRO CORNUCOPIA, #53, May, 1990

"dynamic programming style." In other
words, we want to determine the condi-
tions (i.e., the number and type of ob-
jects) of a program at runtime rather than
at compile time. By delaying system deci-
sions, we allow new discoveries and ad-
justments to be made after we've
"finished" the system.

Not only does this allow us to modify
a program without reprogramming, but
it also allows the user to discover things
about the system the programmer never
even dreamed of (which may work
themselves back into a future version of
the program).

Whether designers like it or not, all
programs go through cycles of design,
use, redesign, reuse, redesign, etc. This
(we maintain) is one of the reasons ex-
isting design techniques don't work so
well — they provide good structure and
documentation (at least theoretically),
but not good design flexibility.

In mainO (Figure 2), you can see that
the user specifies the number of snakes,
the size of the snakes, and the speed of
the simulation by entering parameters
on the command line.

After the ANSI C random number
generator is seeded using the current
time, the arguments are picked off the
command line and an array of pointers
to snakes is created on the heap. (Astute
readers will notice I never free this
space — sloppy, but fairly safe since the
program releases the space when it
exits.)

The #ifdef statements determine
whether the sizes of the snakes are
chosen randomly or are a fixed size
(both options use size_factor).

I create a delay function (called
pauseO defined just before mainO) using
the ANSI C time functions declared in
time.h (see Figure 3).

Get Distracted

I had a lot of fun playing with this
program — got mesmerized actually. If
nothing else, it's a terrific way to waste
CPU cycles.

The source code and SNAKE.EXE are
available through the usual Micro C
channels. The source is also part of the
Using C++ source-code disk (see the
Revolution2 ad this issue).

Farewell

I wrote my first article for Micro C
four years ago. I expected to be
swamped with consulting offers. I
wasn't, but I wrote for every issue since.

Eventually, I combined the articles
into a self-published book, Computer In-
terfacing with Pascal & C. (Academic
Press will publish a very revised and ex-
panded version called PC Interfacing with
C & C++ this fall). Next came Using C++,
and soon you'll see The Tao of Objects co-
written with our pal Gary Entsminger.

Writing for Micro C opened the doors
to other magazines (I'm the C++ editor
at The C Gazette), and speaking at SOGs
gave me the practice and nerve to speak
at "real" conferences. Most important, I
met Larry, Gary, and the gang at Micro
C. No question: writing that first article
was one of the best moves I ever made.

I started in the "brown wrapper"
days ("Hmmm... is this Micro C or
Naked Volleyball Quarterly?"). The ma-
gazine has vastly improved in look and
style, and it seems like we've just hit our
stride. I'm sad to see it go, but perhaps
we've finished creating it. Maybe it
would have become an Institution rather
than a platform for experimentation
(Micro C never did figure out what it was
about....).

There are all kinds of reasons that
Micro C shouldn't have happened: it
wasn't commercial enough, not enough
market research, blah, blah, blah. It did
happen though — a rare and magical
fractal-chaotic event. It happened be-
cause we wanted it to happen, and more
important, because you wanted it to hap-
pen.

Farewell, and look for me in the com-
puter mags....

Bruce Eckel is the author of Using C++
(Osborne /McGraw-Hill, 1989) and a mem-
ber of the ANSI C++ committee. He's the
owner of Revolution!, a firm specializing in
C++ training and consulting.

Reference

Sethi, Ravi; Programming Languages —
Concepts & Constructs (Addison-Wesley,
1989).

♦ ♦ ♦

v>H r
Source Code

from Bruce Eckel's Using C+ +
(Osborne/McGraw-Hill 1989.
ISBN 0-07-881 522-3). Almost
500K: complete source from the
book plus additional projects.
Tested with Zortech C++ 2.06
& Glockenspiel C + + (Common-
view). All chapters and projects
in separate subdirectories , each
with a "makefile."

Projects include:

» Dynamically-sized arrays
(DynArrays)

» Simple Database

» TAWK: A Database Inter-
preter

» A Time-Based Control Sys-
tem

» Simulation Example

» Object-Oriented Menu Sys-
tem

» Text-screen Windows

» Mathematical matrix class

» CAD demo: mouse creates
and moves graphics objects

» MS-DOS directory-manage-
ment class

» Graphic "shape" objects

» Code used to discover com-
piler bugs

And Much More!

To Order, send $25 check to
Revolution2 (address below).
Overseas orders please add $7
for air mail, ana send a U.S.
check in U.S. funds or an inter-
national postal money order.

Revolution2 provides on-
and off-site C++ consulting
& training, and embedded
systems development ser-
vices including analysis,
design, implementation and
desktop-published technical
documentation.

Revolution2

Bruce Eckel Box 760

Kennett Square, PA 19348

(215)444-0828

bix:Beckel cis:72070,32256
net: 72070.3256@compuserve.com

MICRO CORNUCOPIA, #53, May, 1990 25

ABBJBfi B B HBBBi B B

Two-chip Terminal

Motorola's 68HC705 Makes It Simple

Karl builds the world's simplest terminal
around one of Motorola's newest controller
chips.

"his battery-powered RS-232 termi-
nal has it all; it's cheap, easy to
build, uses readily-available parts,
and boasts the newest microcontroller
technology. With this project, you can
get started in 68HC705 design while you
build a versatile piece of computer gear.

The terminal's features make it ideal
for remote and low-power use. The LCD
readout and CMOS microcontroller
(MCU) keep the current drain to below
40 mA. Use of a special-purpose level-
shifter IC allows RS-232 operation from
a single 5 Volt supply.

The 68HC705's on-chip serial port
and timer system help the terminal run
at 9600 baud while controlling the dis-
play and monitoring the keyboard. The
design is flexible enough to allow parts
substitution, if you can't find exactly
what I used in my project.

The Display

My screen is a two-line liquid crystal
display (LCD). You can find many for-
mats of LCDs in most of the large mail-
order ads. Although this project uses a
2-line by 16-character display, feel free to
use anything else you like. Just be sure
your display is compatible with the one
shown.

How do you determine compati-
bility? Look for an LCD that uses the Hi-
tachi HC44780 LCD controller chip. This
is a surface-mount technology (SMT)
chip, soldered onto the back of the dis-
play. It usually appears with one or
more Hitachi HC44100 driver chips (also
SMT).

The HC44780 chip does more than
handle character generation. It contains
an 80-byte buffer. LCDs built with this

chip can therefore display only a maxi-
mum of 80 characters.

But for many applications, a 2-line by
40-character (or 4-line by 20-character)
display will do. By shopping around,
you can find several different formats of
these LCDs. Examples include the Hita-
chi H2750 (1 row of 16 characters), Hita-
chi LM032L or Optrex DMC20215 (2 by
20), or Hitachi LM044L (4 by 20).

Mail-order suppliers known to have
carried LCDs recently include Alltronics,
TimeLine, and Digi-Key. Also check
your favorite surplus house; these dis-
plays are becoming common surplus
items.

Most LCDs include some form of
backlighting. The most common style
uses a strip of electroluminescent mate-
rial (ELM), mounted behind the panel.
When you apply the proper AC voltage
to the two pins on the ELM, it emits a
soft blue or green light. This makes your
terminal easy to use at night or in low
light.

I added a special, three-terminal volt-
age converter designed for use with
ELM. This module changes 5 Volts DC
to 200 Volts AC at about 500 Hz, giving
my display a soft blue backlight.

You might have to scrounge pretty
hard to come up with such a converter;
mine came off a discarded laser-tag ar-
cade game. Half the size of an ice cube,
the converter's case carries the part
number NEL-D32-45, but no manufac-
turer's name.

I have also seen a similar part offered
in the Digi-Key catalog. The terminal
works fine without the converter, of
course, but backlighting adds a nice
touch.

The Keyboard

I chose a surplus Cherry keyboard
(very old, it uses TTL chips). This unit
provides a 15-pin connector for hooking
up to a computer. When you press a key,

the ASCII code associated with that
character appears on seven pins of the
connector, while a strobe pin goes from
ground to +5 Volts, signaling there's
valid data.

This means I only need ten wires to
hook up my keyboard; seven data bits,
one strobe line, power, and ground. I
can use a single eight-bit input port to
handle all my keyboard data. (Of course,
the TTL chips make this keyboard a real
power hog; oh well, it only cost $4.)

With this arrangement, I get all the
control codes, all upper-case letters,
special characters, and numbers; no
lower-case and no PC function-key
sequences. Still, it'll serve my purpose.

There are always other options. For
example, you could use a surplus, unen-
coded switch matrix. This type of key-
board consists of a grid of keyswitches,
with the wires tied to each row and
column brought out to a connector.
Pressing a key shorts a specific row to a
specific column; your software just scans
each row (or column) until it detects a
closed circuit. You can then use the row
and column where the short occurred to
determine which key was pressed.

Though this sounds like extra soft-
ware (it is), the scanned matrix keyboard
offers some benefits. Since the matrix
consumes no power, it makes a perfect
choice for low-power applications. The
surplus market seems glutted with a
variety of unencoded keyboards. If you
shop around, you will surely find just
the right size, shape and style. And the
price will be right; I picked up a brand-
new unencoded keyboard from United
Products for just a few dollars.

You could also opt for a PC-style key-
board. The Keytronics 101-style key-
boards are usually available; I recently
saw an ad from TimeLine that offered a
new XT keyboard for $15.

Unlike the other two keyboard sys-
tems, the PC units provide data for a

26 MICRO CORNUCOPIA, #53, May, 1990

By Karl Lunt

2133 186 th PL, SE

Bothell,WA 98012

(206)483-0447

pressed key via a serial line. A corre-
sponding clock line tells the host com-
puter exactly when each bit of the data
packet is valid. (Check out Bill Curlew's
article, "Building An IBM PC Keyboard
Translator/' in the February/March
1990 issue of Circuit Cellar INK, for
details on PC keyboards.)

This synchronous transmission sys-
tem means the keyboard can send (and,
in some cases, receive) data over just
three wires. At first glance, none of
IBM's three available formats will work
without additional logic. They just aren't
compatible with the 68HC705's syn-
chronous serial line (Motorola calls this
port the serial peripheral interface, or
SPI).

The problem lies with the number of
bits transmitted per keypress. Mode 1
(the oldest and simplest format) uses
eight data bits and one start bit for each
transmission. The other two modes use

additional bits and support bidirectional
transfers.

The SPI, however, can only accept a
transfer of exactly eight data bits; it does
not need nor expect a start bit. To hook a
PC keyboard to the SPI, you could add a
serial-to-parallel converter to the
68HC705. Or, you could put extra cir-
cuitry inside the keyboard to turn the
nine bits into two eight-bit packets.
Neither way was acceptable (I wanted
this project to take only two chips), so
the PC keyboard was out.

As I look that last paragraph over, I
realize there is (as always) another way.
Hooking the keyboard's clock line to an
input port line and the keyboard's data
line to the MCU's TCAP (timer capture)
line would do the trick nicely.

Following a keypress, the data line's
start bit going high-to-low would cause
an interrupt and transfer control to the
TCAP server routine. This routine

ith this project,
you can get started
in 68HC705 design
while you build a
versatile piece of
computer gear.

samples the clock line on the input port,
looking for a low-to-high transition.
Every time such a change occurs, the
routine reads the level on the TCAP line
to get the value for that bit-time. The
server could then construct the keycode,
wait for the stop bit to appear, and exit.

Okay, so you can add a PC keyboard
to this project and still use only two
chips.

The Maxim MAX232 Chip

I've built a couple of projects with
this chip and love using it. Being able to
run RS-232 from a single +5 Volt source
reduces circuit complexity, chip count,
and board size.

The Maxim chip works by generating
+8 and -8 Volts on-chip from the sys-
tem's +5 Volt supply, using a charge-
pump technique. You hook up four 22
mfd capacitors for the charge-pump; the
chip gives you two output and two
input buffers that are RS-232 compatible.

You might run into some problems,
however. The chip does not supply a lot
of current to the RS-232 buffers. If you
try to push 9600 baud over a long cable,
you could get errors.

A new version of this chip, the

MICRO CORNUCOPIA, #53, May, 1990 27

MAX233, puts the capacitors on the
chip. This will further reduce the parts
count for your RS-232 interface. But this
chip suffers from the same low drive
current as its MAX232 brother. At least
one engineer I know has cautioned me
about using either chip to talk distances.
Having said this, I'll add that I have
yet to see any data errors directly due to
the Maxim devices. Plus, both enjoy a
wide following in the industry. I cer-
tainly intend to keep using them.

The MCU

Consider what it takes to make an
RS-232 terminal. You need a bi-
directional serial port good for 9600
baud and plenty of I/O lines to hook up
to your LCD and keyboard.

You also need a timer system for
properly clocking the control signals into
the LCD, an interrupt system for catch-
ing the high-speed data flow on the se-
rial line, and enough RAM and EPROM
to hold your program.

Motorola's 68HC705 provides all this,
at a current drain of less than 10 mA.
Check out Micro C Issue #49 for an in-
depth look at the 68HC705. For now, I
will only discuss those aspects of the
MCU needed by the two-chip terminal
project.

You control the extensive array of
I/O and timing functions on the MCU
by writing 8-bit values into specially as-
signed memory locations, called regis-
ters. For example, you can use all eight
bits of I/O port A as outputs by storing
the value $FF into location $04, known
as DDRA (data direction register A).

By using a 4.0 MHz crystal, the MCU
can hit 9600 baud on the serial port. Al-
though the project currently does not
support it, you could easily change baud
rates from the keyboard or a row of dip
switches. To alter baud rates, simply
write different timing values to two
baud-rate selector registers.

You must wait a prescribed amount
of time after each change of the LCD. I
use the MCU's 16-bit timer system to
generate 200 usee time slices. The LCD
routines simply wait until the correct
number of slices (called tics) elapse
before sending the next command.

Since the terminal could receive a
stream of 9600 baud data, the MCU
must handle the incoming data immedi-
ately. The MCU's serial communication
interface (SCI) can generate an interrupt
upon receiving a character from the se-
rial port.

Figure 1 — LCD and Keyboard Interface to '705

68HC705

PA0

PA1

PA2

PA3

PA4

PA5

P1
1
2
3
4
5
6
10

LCD

4 MHz

DB4

0B5

DB6

DB7

GND

,.R/W

Vee

Vcc

+5V

4 * 22pF

RESET

+5V

+
=F2.2ljF

1RTJ

10K

+5V

PD1

PD0 9

><}

MAX232

2 0_D>°7_

RS-232
■*-3

+5V

CM
X

7805

FRONT
VIEW

IN| | |0UT
GND

_+
"10IJF

IN4001

+5V

9-30V DC

7805

500ljF

GND

50ljF

68HC705

PB0

12
13
14
15
16
17
18
19

PB1

PB2

PB3

PB4

PB5

PB6

PB7

P2
1
2
3
4
5
6
7
8

9
10

KEYBOARD

BIT

(15)

BIT 1

(14)

BIT 2

(13)

BIT 3

(12)

BIT 4

(10)

BIT 5

01)

BIT 6

(9)

STROBE

(6)

+5V

(4)

GND

(8)

28 MICRO CORNUCOPIA, #53, May, 1990

End diskette compatibility problems.
Call Emerald Microware.

CompatiCard I by Micro Solutions

This four drive universal floppy controller will let you
run up to 16 disk drives (4 per CompatiCard), including
standard 360K, 96 TPI, high density 1 .2M, 8" (SSSD
or DSDD), and 720k/1.44M 3V2" drives. The Compati-
Card I comes with its own MS-DOS driver, utility pro-
grams, and will let you boot on an XT (must be used
as a secondary controller on an AT or 386). Use it
with UniForm-PC for maximum versatility.

CompatiCard I Board $ 1 19.95

CompatiCard I with UniFORM-PC $ 179.95

8" drive adaptor board $ 15.00

External drive cable set $ 15.00

CompatiCard II by Micro Solutions

Two drive version of the CompatiCard, for the XT or
AT. Same drive support as the CompatiCard I except
no 8" or single density.

CompatiCard II $ 89.95

••• Special ••• CompatiCard II with

internal 1.2M or 1.44M drive $ 199.95

CompatiCard IV by Micro Solutions

Meet the newest four drive controller in the Compati-
Card family. This CompatiCard may be used as a
primary or secondary controller in almost any PC, AT,
or 386 System. Boot or use 360k, 720k, 1 .2M, 1 .44M,
or 2.88M, at any location in your system. The Com-
patiCard IV has a BIOS ROM on board so no external
driver software is required.
CompartiCard IV $ 139.95

Megamate by Micro Solutions

You don't have to be a computer expert to install this
attractive 3V2" external drive on your PC or AT. Just
plug the MegaMate controller board into any empty
slot, attach the drive cable, run the installation soft-
ware, and you're ready to run 720k or 1.44M diskettes.
Megamate $ 329.95

Apple M MS-DOS

MatchPoint-PC by Micro Solutions

Apple II or NorthStar diskettes in your IBM? The
MatchPoint-PC board for the PC/XT/ AT works with
your standard controller card to let you read and write
to NorthStar hard sector and Apple II diskettes on
your PC. INCLUDES UniForm-PC program, as well
as utilities to format disks, copy, delete, and view files
from Apple DOS, PRODOS, and Apple CP/M disks.
MatchPoint-PC Board $ 179.95

VISA

/ ^

UniForm-PC by Micro Solutions

Have you ever needed to use your CP/M diskettes
on your PC? Now you can access your CP/M files
and programs on your MS-DOS computer just as you
would a standard MS-DOS diskette. UniForm allows
you to use standard DOS commands and programs
right on your original diskette without modifying or
copying your files. UniForm-PC allows you to read,
write, format, and copy diskettes from over 275 CP/M
and MS-DOS computers on your PC, XT, AT, OR 386.
With UniForm-PC and the CompatiCard, you can use
5V4" high density, 96TPI, 3V 2 " (720k/1.44M), and
even 8" drives.

UniForm-PC by Micro Solutions $ 64.95

Also available for Kaypro, & other CP/M computers

CP/M • MS-DOS

UniDOS Z80 Coprocessor Board

by Micro Solutions

Don't throw out all of those old, reliable CP/M pro-
grams, run them at LIGHTNING speed on your PC
or AT with the UniDOS 8MHz. Z80 coprocessor
board. And the UniDOS Z80 runs so smoothly and
transparently that you won't even be able to tell
whether you're running DOS or CP/M. UniDOS emu-
lates most common computers and terminals such
as Kaypro, Xerox 820, Morrow, Osborne, VT100, and
many others. Supports all standard CP/M system
calls, and now works with MS-DOS version 4. In-
cludes UniForm-PC.
UniDOS Z80 Coprocessor Card $ 169.95

UniDOS

by Micro Solutions

If you have a fast machine or have a V20 chip in-
stalled, you may not need to use a card slot to run
your CP/M programs. Run 8080 code directly on the
V20, or use emulation mode for Z80 programs.

UniDOS by Micro Solutions $ 64.95

UniDOS w/ UniForm &V20-8 chip $ 135.00

MatchMaker by Micro Solutions

Now you can copy your Macintosh diskettes right on
your PC/XT/AT with the MatchMaker. Just plug your
external 3V2" Macintosh drive into the MatchMaker
board and experience EASY access to your Mac disk-
ettes. Includes programs to read, write, initialize, and
delete files on your single or double sided Mac disks.

MatchMaker Board $ 139.95

MatchMaker w/External Mac Drive ... $ 325.00

Copy II PC by Central Point Software

Don't let a damaged copy protected diskette stop you
cold. Copy II PC lets you back up your master disks so
you can keep going even when your key disk can't.
Copy II PC $ 24.95

Copy II PC Deluxe Option Board

by Central Point Software

Have a copy protected diskette with a particularly
stubborn protection scheme? Would you like to be
able to read your Macintosh disks in your 3V2" in-
ternal drive in your PC or AT? Repair a disk that's
damaged, even between sectors? How about speed-
ing up your hard disk backups (if you are using PC
Tools Deluxe)? The Copy II Deluxe Option Board can
help you do all of this, and more. A must for the
sophisticated user.
Copy II Deluxe Option Board $ 139.95

CentralPoint

Software

J INCORPORATED

PC Tools Deluxe V5.5

by Central Point Software

This is one of the great bargains in MS-DOS utility
software. But with so many features built in, we think
that many people are overlooking the REAL value
in PC Tools: DISASTER RECOVERY! Sure, the shell
is great for copying, viewing, editing, and deleting
files, and the desk top environment is nice for its
appointment calendar, note pad, phone dialer, calcu-
lators, and ASCII table. The fast hard drive backups
can't be beaten by any other program, and the file
unfragmenter speeds up hard drive accesses. But
where else can you get all of that along with UN-
DELETE, REBUILD, and UNFORMAT? Have you
ever entered "ERASE *.*" and realised when all was
said and done that you were in the wrong directory?
How about those dreaded messages from CHKDSK,
like "File allocation error...."? The very first time you
recover your missing files will make you a believer.
Don't wait until it happens.
PC Tools Deluxe V5.5 $ 99.95

EMERHLD
MICROHRRE

P.O. Box 1726

(503) 641-8088 Beaverton, OR 97075

Call or write for our complete catalog of

software, parts, accessories and

complete repair services for the

Kaypro, Xerox 820, and IBM PC/AT.

VISA and Mastercard accepted. Please include $6.00
shipping and handling, $8.50 for COD, UPS-Blue or
RED Label additional according to weight. Prices
subject to change without notice. Please include your
phone number with all correspondence.

Reader Service Number 10

MICRO CORNUCOPIA, #53, May, 1990 29

I use the SCI's interrupt to grab each
character as it comes in, then stash the
data in a circular queue. Eventually the
mainline code gets around to pulling the
character from the queue and processing
it.

Data from the keyboard arrives much
more slowly. (In my somewhat ad-
vanced years, I can no longer type much
above 4800 baud.) The MCU simply
glances from time to time at the key-
board port. If the line tied to the key-
board's strobe signal shows that data is
available, the MCU grabs it.

Data from the terminal to the host
computer also uses the SCI, but the pro-
gram has no need to send anything in a
steady stream. It (currently) sends a
single character to the computer each
time a key is pressed, so this function
hardly requires interrupts.

On To The Software

I've divided the assembly language
code into several major functions. The
LCD routines initialize and update the
display. The output compare (OCMP)
server handles the 200 usee tic inter-
rupts. The SCI server takes care of in-
coming data from the computer. The
mainline code does everything else, such
as processing keyboard data and keep-
ing the display looking pretty.

I based my LCD library on informa-
tion from Ed Nisley's article, "The True
Secrets of Working with LCDs," from
the April/May 1989 issue of Circuit Cel-
lar INK. Ed's explanation of the LCD's
inner workings really helped.

Because he did all his software in C,
Ed didn't concern himself with the re-
quired delays between LCD commands;
the C code ran slow enough all by itself.

Adding Delays

Writing in assembly language, how-
ever, means adding delays. These waits
can be significant. The initialization
ritual, for example, calls for delays as
long as 5 msec; the shortest delay (be-
tween sending two visible characters to
the display) still takes 120 usee. (Note:
All LCD commands have a required
minimum delay before a subsequent
command will be correctly processed.
However, there is no maximum delay.)

I solved this delay problem with the
OCMP server and a global variable
called WAIT$.

The MCU's 16-bit timer continually
counts down from $FFFF, through
$0000, and starts over again at $FFFF.

The output compare register (OCR) can
hold an arbitrary 16-bit value. When the
value of the free-running timer matches
the value stored in the OCR, the MCU
generates a timer interrupt.

To generate a fixed-length time slice,
or tic, simply figure out how many
timing counts correspond to the delay
you need, add that value to the current
value of the timer, and store the sum in
the OCR. After the proper number of
counts, the timer will reach the value
you stored in the OCR and trigger an in-
terrupt. Your interrupt service routine
then calculates the sum for the next tic,
updates the OCR. . . .

In my case, I needed to count off the
correct number of 200 usee tics. So my
interrupt routine does more than just set
up the next interrupt count. It also
checks the value in WAIT$; if that value
is not yet zero, the routine decrements it.

Therefore, any routine that must wait
(for example) 400 usee simply loads a
value of 2 (to wait for two tics) into
WAIT$, then hangs around until WAITS
becomes 0.

You will notice my software only
uses the A and X registers. That's all you
get! The tiny programming model on
this chip really makes you work to get
the most out of your code.

More About LCDs

I connected my LCD to the terminal
board using a 4-bit interface. Since the
LCD interface also requires a register
select line (RS) and an enable line (E), I
could run the whole display with only
six I/O lines. Refer to the schematic for
details. Pay particular attention to the
data lines. Note that the low four bits of
the MCU's I/O port connect to the high
four data bits on the LCD.

Unfortunately, the LCD comes out of
reset in an 8-bit interface mode. You
must follow a carefully defined ritual in
setting the LCD to read the 4-bit inter-
face. The code in LCDINIT does the job.

First off, I reset all the lines to the
LCD and load a 15 msec power-up delay
into WAITS. I then write a sequence of
three commands, each resetting the LCD
to an 8-bit interface. (Yes, I know we
want a 4-bit interface, but you gotta do
this just like the book says, or it isn't
going to work.) Note the different delays
used following each command.

Next, I set up the display for a 4-bit
interface. The remaining commands fix
the type of display format, cursor con-
trol, font, and cursor address. If you

want to try other display characteristics,
check the Hitachi manual or Ed's article
in Circuit Cellar INK for details.

The LCD accepts two types of data.
Commands, such as those discussed
above, must be written to the display
with the RS line low. Data, such as dis-
played text, must be sent while RS is
high. To make these operations easier, I
included LCDCMD and LCDCHAR.

These routines use a common section
of code, found at LCDCHAR1. Each sets
up the RS line as needed, loads an ap-
propriate delay value into the AR, and
drops into the code at LCDCHAR1.
From here, the byte gets divided into
two 4-bit nybbles. Calls to LCDOUT
then transfer the nybbles (MSB first) to
the display.

LCDSTRING simply sends each char-
acter of a null-terminated string to the
display. The only tricky bit involves the
load indexed, indirect addressing mode
of the 68HC705; the chip doesn't have
such a mode.

To provide this essential addressing
mode, I resorted to self-modifying code
(actually, a RAM-based subroutine,
which is just as bad).

The initialization code for the termi-
nal program builds up a two-instruction
subroutine called LDAIND by writing
two opcodes into low RAM. Address
LDAIND holds a $D6 (load accumula-
tor, indexed, with a 16-bit offset), while
address LDAIND+3 holds a $81 (return
from subroutine). The two bytes at
LDAIND+1 and LDAIND+2 get
changed by another routine before cal-
ling LDAIND.

At run-time, a section of code stores
the 16-bit address of a null-terminated
string into the two bytes at LDAIND+1.
It clears the XR (X index register), then
calls LCDSTRING. Since the LDAIND
subroutine is all set up, LCDSTRING can
simply call the routine to get the charac-
ter in the string pointed to by the XR. It
then increments the XR, gets the next
character, etc., until it gets the null mark-
ing the string's end.

And Now The Hardware

I built the prototype terminal on a
Radio Shack experimenter's card (276-
158A). You can find nearly all the parts
in a Radio Shack or Active Electronics
catalog; write these companies for a cat-
alog if you don't have one already.

The only part you might have trouble
with is the 68HC705 MCU. You will
need to contact a Motorola distributor

30 MICRO CORNUCOPIA, #53, May, 1990

C CODE FOR THE PC

source code, of course

MS-DOS File Compatibility Package (create, read & write MS-DOS file systems on non-MS-DOS computers) $750

CQL Query System (SQL retrievals on B-trees plus windows) $325

Graphic 5.0 (high-resolution, DISSPLA-style scientific plots in color & hardcopy) $325

PC Curses (Aspen, Software, System V compatible, extensive documentation) $290

C-Data Manager (object-oriented data management, persistent objects from runtime definitions, network and entity models) $250

MEWEL (extensible window and event library by Magma Software; message-passing & object-oriented; SAA-compatible; dialog editor) . . $250

TurboTeX (Release 2.0; HP, PS, dot drivers; CM fonts; LaTfiX; MetaFont) $250

db.File & db.Retrieve by Raima (B-tree and network database with SQL query and report writer; multi-user $475) $245

Greenleaf Communications Library (interrupt mode, modem control, XON-XOFF; specify compiler) $225

CDirect (multi-user hashed file manager; variable length fields, binary or ASCII data, alternate keys) $210

SilverComm (complete asynchronous communications library) $210

Wendin-DOS Plus (self-bootable, multitasking, multiuser MS-DOS replacement; includes XTC editor) $180

QuickGeometry Library (large collection of mathematics, graphics, display & DXF subroutines for CAD/CAM/CAE/CNC) $170

CBTree (B+tree ISAM driver, multiple variable-length keys) $165

TlirboGeometry (library of routines for computational geometry, Version 3.0) $160

AT BIOS Kit (roll your own BIOS with this complete set of basic input/output functions for ATs) $160

WKS Library Version 2.01 (C program interface to Lotus 1-2-3, dBase, Supercalc 4, Quatro, & Clipper) $155

NEW! C Generator (generates C code to read & write file records defined with C structure syntax) $150

OS/88 (industrial-strength U«*x-like operating system, many tools, cross-development from MS-DOS) $150

Cephes Mathematical Library (over 100 high-quality, double-precision scientific functions) $150

ME Version 2.1 (programmer's editor with C-hke macro language by Magma Software; Version 1.31 still $75) $140

Vmem/C (virtual memory manager; least-recently used pager, dynamic expansion of swap file) $140

Turbo G Graphics Library (all popular adapters, hidden line removal) $135

Rogue Wave Vector & Matrix Classes (inc. C++ overloadings for standard operators, matrix inversion & FFT; Zortech or GNU C++) . . $125

Power Search by Blaise Computing (regular-expression compiler, generates machine code on the fly) $120

Install 2.3 (automatic installation program; user-selected partial installation; CRC checking) $120

TE Editor Developer's Kit (full screen editor, undo command, multiple windows) $115

NEW! Hold Everything (spawn new programs; swap parent to EMS or disk; handles video, interrupts, & environment; returns error level) $105

B-Strings (dynamic string handling; cut, copy, paste, search, user input, etc.; non-fragmenting memory management) $105

Minix Operating System (Version 1.3; U**x-like operating system, includes manual) $105

PC/IP (CMU/MIT TCP/IP for PCs; Ethernet, Appletalk & NETBIOS drivers, RVD, gateways) $100

B-Tree Library & ISAM Driver (file system utilities by Softfocus) $100

The Profiler (program execution profile tool) $100

QC88C compiler (ASM output, small model, no longs, floats or bit fields, 80+ function library) $90

Booter Toolkit (floppy disk bootstrap routines, DOS file system, light-weight multitasking, windows, fast memory management) $85

Otter 1.0 (beautiful theorem-prover by Bill McCune; includes manual & two books by Wos; complete starter kit) $80

JATE Async Terminal Emulator (includes file transfer and menu subsystem) $80

PowerSTOR(upto a gigabyte of heap space on extended memory, expanded memory, and/or hard disk) $80

MultiDOS Plus (DOS-based multitasking, intertask messaging, semaphores) $80

HY-PHEN-EX (a hyphenator for American English with over 4,800 rules) $75

Make (macros, all languages, built-in rules) $75

evalO (C function to evaluate ASCII infix expression string; 17 built-in functions) $75

XT BIOS Kit (roll your own BIOS with this complete set of basic input/output functions for XTs) $75

Professional C Windows (lean & mean window and keyboard handler) $70

Heap Expander (virtual memory manager using expanded memory, extended memory, and disk space) $65

Quincy (interactive C interpreter) $60

SymtabflMree (general-purpose symbol table/parse tree construction and management package; specify Symtab or Ptree) $60

Coder's Prolog (Version 3.0; inference engine for use with C programs) $60

Async-Termio (Unix V compatible serial interface for MS-DOS; stty, ioctl, SIGINT, etc.) $55

Backup & Restore Utility by Blake McBride (multiple volumes, file compression & encryption) $50

Floppy TAR (TAR backup and restore on MS-DOS devices; direct access to non-standard devices) $50

SuperGrep (exceptionally fast, revolutionary text searching algorithm; also searches sub-directories) $50

OBJASM (convert .obj files to .asm files; output is MASM compatible) $50

Multi-User BBS (chat, mail, menus, sysop displays; does not include Hayes modem driver) $50

La placeB (LaPlace polynomials, real and complex) $50

CLIPS (rule-based expert system generator, Version 4.3; advanced manuals available) $50

Pascal P-Code Compiler & Interpreter (full ISO standard Pascal) $50

PCHRT (40 functions to manage multiple microsecond timers; generate precision delays; insert timers on any interrupt) $45

Kier DateLib (all kinds of date manipulation; translation, validation, formatting, & arithmetic) $45

Fortran-to-C Translator by Polyglot (Fortran-IV-like Fortran to ugly C; plan to adapt to your own flavor of Fortran) $40

DES Encryption & Decryption (2500 bits/second on 4.77 MHz PC for on-the-fiy encryption at 2400 baud) $40

FlexList (doubly-linked lists of arbitrary data with multiple access methods) $40

Virtual Memory Manager by Blake McBride (LRU pager, dynamic swap file, image save/restore) $40

Heap I/O (treat all or part of a disk file as heap storage) $40

Bison & BYACC(YACC workalike parser generators; documentation; no restrictions on use of BYACC output) $35

PC-XINU (Comer's XINU operating system for PC) $35

RXC& EGREP (Regular Expression Compiler ancf Pattern Matching; RXC makes finite state machine from regular expression) $35

Cheaper! REGX Plus (search and replace string manipulation routines based on regular expressions) $30

CCALC (handy extended-precision calculator; real and complex models; many built-in functions) $30

GNU Awk & Diff for PC (both programs in one package) $30

6-Packof Editors (baker's half-dozen public domain editors for use, study & hacking; includes microEmacs 3.10 & Stevie, a vi clone) .... $30

Crunch Pack (14 file compression & expansion programs) $30

NEW! PC-MAIL (UUCP mailer by WietseZVenema; send, receive, and manage UUCP mail) $25

FLEX (fast lexical analyzer generator; new, improved LEX; official BSD Version 2.1 with docs) $25

List-Pac (C functions for lists, stacks, and queues) $25

Using C++ Library (the code from the book by Bruce Eckel and then some; Zortech 2.0 compatible) $25

A68 (68000 cross-assembler) $20

XLT Macro Processor (general purpose text translator) . $20

Data

Moby Pronunciator (150,000 words & phrases encoded with full IPA pronunciation & emphasis points; 900 distinguished by part-of-speech) . $160

Moby Part-of-Speech (200,000 words and phrases described by prioritized part(s)-of -speech) $120

Moby Hyphenator (150,000 words fully hyphenated/syllabified) . . $105

Moby Words (500,000 words & phrases, 9,000 stars, 15,000 names) $65

Smithsonian Astronomical Observatory Subset (right ascension, declination, & magnitude of 258,997 stars) $60

U.S. Cities (names & longitude/latitude of 32,000 U.S. cities and 6,000 state boundary points) $35

The World Digitized (100,000 longitude/latitude of world country boundaries) : $30

KST Fonts (13,200 characters in 139 mixed fonts: specify TfeX or bitmap format) $30

Interactive Computer Ephemerii (high-precision moon, sun, planet & star positions; USNO (no source) & Downey 4.8 (C source)) $30

The Austin Code Works Voice: (512) 258-0785

11100 Leafwood Lane info@acw.com BBS: (512) 258-8831

Austin, Texas 18150-3409 USA FAX: (512) 258-1342

Free surface shipping for cash in advance For delivery in Texas add 7% MasterCard/VISA

Reader Service Number 4

MICRO CORNUCOPIA, #53, May, 1990 31

Upgrade!
NEW!

directly for this chip, as I don't think it is
in the mail-order houses yet. Almac
Electronics (206-643-9992) in Bellevue,
Washington, will handle the parts mail-
order in singles; price for the -S version
of this part (with the EPROM window)
is about $18.

Be sure to get a copy of Motorola's
HC705 C8 Technical Data book, literature
number MC68HC705C8/D. Besides all
the technical info on using the 'HC705,
this book includes a full schematic for
building a programming board (piece o'
cake). It also includes instructions on
using the programming board.

You can use just about any wiring
technique you want for this project. I
used point-to-point soldering with 30
AWG wire; wirewrapping would work
also.

By the time you read this, I will have
a general-purpose experimenter's
printed wiring board (PWB) for the
68HC705. The board will contain all the
basic circuitry needed to build a work-
ing MCU project, and could easily be the
only PWB you might need to develop an
application.

For example, this two-chip terminal
project would consist of a working ex-

STOMP OUT
EPROM MADNESS

The PROM KING emulates EPROMS, saving
both time and money during your development
cycle. Programmable in seconds via your PC
printer port or any computer RS232 port, it can
emulate most 27xxx devices.

• 8K-8M bit devices • 8-256 bit downloads

• High speed download: • Easily expandable:
-Universal RS232 -4 EPROMS per unit
-PC printer port -Up to 8 units

• Menu driven software • Also programs like

• Battery backup a real EPROM

$599 for 150nS units with 256K bits
Ask for pricing of other options

Made in USA by:-

TmXELWBSINC

BOX 239 • RONKONKOMA.NY-1 1 779
51 6 -737' 5147

Reader Service Number 1 78
32 MICRO CORNUCOPIA, #53, May, 1990

perimenter's PWB connected to a key-
board and LCD.

After you get your terminal wired up,
you need to assemble the source code. I
use a PC/XT clone running Motorola's
AS5 Freeware cross-assembler. Bor-
land's Sidekick handles my editing, and
I burn the finished code into the MCU
using Motorola's M68HC05PGMR pro-
grammer board and PROG7 support
software.

You can download both PROG7 and
AS5 from the Motorola Freeware BBS
(512-891-3733; 8 data, no parity, 1 stop).
Motorola uses the BBS to distribute free
cross-assemblers (available for both the
PC and Mac), working source code for
MCU designs, technical updates, and
marketing announcements. Dial up and
take a look.

Even if you don't yet have a key-
board, you can still hook up the display
and apply power (use a DC source from
+7 to +25 Volts). Your LCD should show
the power-up announcement "Work-
ing..."

If you don't get the proper display,
use an oscilloscope or logic probe to
check pin 35 of the MCU. My software
toggles this line each tic; you should see
a 50% duty-cycle square wave that
changes state every 200 usees. If you see
this signal, you likely have an error in
your LCD connection.

If you don't see this signal, check
your board's power supply and oscilla-
tor circuitry; you may not be getting +5
Volts to the chip, or your crystal may not
be oscillating.

After you get the display working,
wire up your keyboard using the sche-
matic as a guide. The software expects
you to supply seven data bits (on port B,
bits 0-6) and a positive-going strobe
(port B, bit 7); you can use any keyboard
that will supply these signals. Note that
the circuitry does not supply +12 Volts;
keep that in mind when you hit the sur-
plus stores.

With your keyboard installed, tem-
porarily short pins 2 and 3 of the RS-232
connector. When you apply power and
press a key, the display should echo that
key. If so, you have a working terminal!

Time To Tinker

Since everything about this terminal
can be found in the software, you can re-
ally customize this project. Consult the
Hitachi manual for the low-level LCD
commands if you want to change the
shape/blink of the cursor. The Maxim

chip has a pair of unused RS-232 buffers;
you could hook these up to I/O pins on
the MCU and provide hardware hand-
shaking on the serial line.

Since a keyboard character goes
through the MCU before it gets shipped
out the serial port, you can easily imbed
function codes or signature keys into the
software.

This design leaves an entire 8-bit port
vacant; how about hooking up relays,
lights, a piezo-beeper (for a control-G
bell code), or other control devices? You
could even hook up an external A/D
converter and have a stand-alone en-
vironmental control station. The station
could then download data over the se-
rial port to a host computer, while the
keyboard lets you run programs on the
host for evaluating data.

Anyone doing field work would
probably find a use for this terminal.
How about hooking up some rechargea-
ble batteries (or a solar panel, if you use
a switch matrix keyboard) and taking
the terminal outdoors?

If you run road rallies, you could
easily connect the terminal (complete
with backlighting) to a small rally com-
puter stashed in the backseat, and do
some real digital navigation.

That's About It

This project was great fun. Every-
thing I develop with the '705 makes me
want to try another design or two. If you
customize this project (as I am sure you
will), I'd like to hear what you have
done.

Active Electronics

P.O. Box 9100

Westborough, MA 01581

(800) 888-9939 (outside New England)

(800) ACTIVE6 (New England)

TimeLine Inc.

1490 W. Artesia Blvd.

Gardena, CA 90247

(800) 872-8878 (outside California)

(800) 223-9977 (California)

United Products
1123 Valley Street
Seattle, WA 98109-4425
(206) 682-5025

♦ ♦ ♦

Two great tods.

4&k ^

SAYWHAT?!

The lightning-fast
screen generator.

Whether you're a novice
programmer longing for
simplicity, or a seasoned
pro searching for higher
productivity, you owe it
to yourself to check out
Saywhat. You see, with
Saywhat, you can build
beautiful, elaborate, color-
coded screens in minutes!
That's right. Trulyfantastic
screens for menus, data
entry, data display, and
help-panels that can be
displayed with as little as one
line of code in any language.

TOPAZ

The breakthrough
DBMS toolkit for Pascal

If you'd like to combine the raw power
and speed of Turbo Pascal or Micro-
soft's QuickPascal with the simplicity s
and elegance of dBASE, Topaz is ^~
just what you're looking for. J^i
That's because Topaz was /

specially created to let you

XT)

enjoy the best of both worlds. /$¥/'
The result? You create com- AvV^-
plete, truly dazzling applicat- yS
ions in a very short time. And no wonder.
Topaz is a comprehensive toolkit of
dBASE -like commands and functions
designed to help you produce
outstanding, polished programs, fast.
Check out these powerful features:

Here's what you get:

• Design screens, windows, and
moving bar menus!

• Easy-to-use, powerful editor lets
you create screens in a jiffy.

• Pop up your screens and menus
with one line of code in dBASE,
all the dBASE compilers, your
favorite BASIC, Pascal, or any
other language!

• Screen Library Manager.

• Generates runtime code.

• No runtime license or royalty fees.

• Comes with a 100 page manual,
plus dozens of sample programs
and free utilities.

$49

ORDER NOW.
YOU RISK NOTHING.

Thousands of satisfied customers
have already ordered from us. Why
not visit your dealer or call toll-free,
right now and put Saywhat and Topaz
to the test yourself? They're fully
guaranteed. You don't risk a penny.

Special limited-time offer! Save $26.
Buy Saywhat?! and Topaz together
for just $99 (plus $5 shipping and
handling, $10 Canada, $25 Inter-
national, Calif, residents add 7%).

Visit your nearest dealer

or call toll-free:

800-468-9273
In California: 800-231-7849
International: 415-571-5019

Software Science, Inc.
100 Valley Drive, Brisbane, CA 94005

rfTis K \ l MONEY BACK GUARANTEE.

/XA/ Sw -V / '' If you aren't completely delighted with Saywhat
/ / c 7v' ' or Topaz, for any reason, return them within

; 30 days for a prompt, friendly refund

Dealers: SAYWHAT?! and TOPAZ are

available from Kenfil Distribution.

and in Europe from

ComFood Software, W. Germany 49-2534-7093

• Over 220 routines all with easy-to-
use, dBASE-like syntax.

• Data entry routines like SAY, GET,
PICTURE, RANGE, color selec-
tion, unlimited data validation.

• Open up to 10 DBF files, with up to
15 indexes with USE, SELECT,
SKIP, APPEND, PACK, INDEX
ON, SET INDEX TO, and FIND.

• No need to buy dBASE. CREATE,
BROWSE and REPORT utilities
included.

• Easily implement Saywhat and
Lotus-style moving bar menus.

• BROWSE or EDIT any DBF file
with just one line of code! Pro-
grammable and windowed too.

• Pick from windowed data or file-
names with one line of code.

• Comprehensive Time & Date
math in 7 international formats.

• Powerful code and report generators
included!

• Comes with a complete 320 page
manual, plus sample programs to
get you started. &HA9S

Guaranteed!

S O F T W

C I E N C E

INC.

Reader Service Number 129

MICRO CORNUCOPIA, #53, May, 1990 33

A Roundoff Roundup

When Rational Numbers Aren't Really Rational

Are rational numbers really rational?
Perhaps not. Does it matter? Most likely.

Gary Entsminger, in his Tidbits
column "Faith in Numbers:
Chaos in Chaos" (Micro C #48,
July- August 1989), discusses the prob-
lem of roundoff error in recursive itera-
tions of non-linear equations. Mr. Ents-
minger points out a problem facing
computer mathematics which, at least
for some applications, has a simple solu-
tion. This article makes several refer-
ences to Mr. Entsminger's column, so
you may wish to reread it.

Mr. Entsminger defines a rational
number as a real number which is either
reducible to an integer, expressible as a
terminating decimal fraction, or express-
ible as a decimal fraction with a repeating
sequence of digits. This is a common and
workable definition, but not an exact one.
Take an ordered pair (x,y) of integers
and define an equivalence relation = such
that for any two ordered pairs (x,y) and
(a,b) with x<a and y<b, (x,y) = (a,b) if and
only if there exists an integer z where az
= x and bz = y. Think of multiplying both
the numerator and denominator of a frac-
tion by the same number. Then define
two binary operations * and + by:

(x,y) * (a,b) = (xa,yb)
(x,y) + (a,b) = (xb + ay,yb)

This set of all ordered pairs of in-
tegers, in conjunction with the relation =
and these two operations, is the number-
theoretic definition of the rational num-
bers. Think of x as the numerator and y
as the denominator of a fraction.

This means I can represent a rational
number as an array of integers. Compare
this to floating point. Longint in Turbo
Pascal is 32 bits long and can represent

2 unique integers. If I use two of these
to represent a rational number, I can rep-
resent 2 33 integer values with (2 33 - l)/2
divisions between each integer. (I cannot
have zero for a denominator, and nega-
tive values in the denominator cancel out
negatives in the numerator.)

I can then represent 2 - 2 numbers,
with some repetitions. Using the same
number of bits in floating point notation,
I can represent 2 different values with
about the same number of repetitions, so
floating point notation can represent a
much larger set of numbers.

There is a tradeoff. No matter how
many bits I use, floating point can never
represent some simple rational numbers
accurately. Take the rational number l /s
for example; there are no integers a and b
for which 16 = a + 2 . The entire range of
numbers represented by floating point
notation is riddled with numbers which
can't be represented exactly, namely any
simplified rational number m/n where n
cannot be represented as a sum of non-
zero powers of two.

Mr. Entsminger states that recursive
iterations of non-linear equations like
Nextx = RX(1 - X) must "...surely en-
counter at least one irrational number."
Since the rationals are closed under mul-
tiplication and addition, this statement, as
written, is false. Even if we consider
roundoff error, this error is due to a finite
number of decimal places, and any termi-
nating decimal fraction is a rational. But I
think Mr. Entsminger is attempting to ex-
press something more complex than fun-
damental concepts of basic algebra.

We call a real number z "rational" if it
can be "rationalized": if we can find in-
tegers x and y such that x/y = z. There is
a similar process which seems natural for
floating point notation. We call a real
number z "rational," as a floating point
number, if we can find integers a and b
for which z = a + 2 . Then, in floating

o matter how
many bits I use,
floating point
notation can never
represent some
simple rational
numbers
accurately.

34 MICRO CORNUCOPIA, #53, May, 1990

By Lance Dannon Bresee

University of California Observatory

Lick Observatory Electronics Lab

University of California at Santa Cruz

1156 High Street

Santa Cruz, CA 95064

Figure 1— RATIONAL.PAS

Unit Rationals ;
Interface

Type

Rational = Array [1.. 2] of longint;

Function Prime ( N : longint ) : boolean ;

{Returns TRUE if integer N is Prime and False otherwise)
Function GCD( A, B : longint): longint;

{Returns the greatest common divisor of A and B}
Function LCM( A, B : longint): longint;

{Returns the least common multiple of A and B)
Procedure Reduce ( Var X : Rational) ;

{Reduces rational X to simple form)
Procedure Invert ( Var X : Rational) ;

{Inverts rational X}
Procedure Radd( XI, X2 : rational; Var Y : Rational) ;

{Returns the reduced sum of XI and X2 in Y}
Procedure Rsub(Xl,X2 : Rational; Var Y : Rational) ;

{Returns the difference of XI - X2 in Y}
Procedure Rmult(Xl,X2 : Rational; Var Y : Rational);

{Returns the product of XI and X2 in Y)
Procedure Rdiv(Xl,X2 : Rational; Var Y : Rational);

{Returns the result of X1/X2 in Y }

Implementation

Function Prime ( N : longint ) : boolean;

Var

Pr, Done : boolean;
Divisor : longint ;
Count : byte ;
Begin {Prime}
Pr := False;
Done := False;
Divisor := 3;
Count := 0;
If N <= 10 then
begin {if}

Done :— true;

If N in [1,2,3,5,7] then Pr := true;
end {if}

Continued on page 36

point notation, V$ is "irrational." Because
of the large number of these "irrational"
numbers, computer science has, by
adopting floating point notation for all
non-integral numbers, wandered up
something of a blind alley.

Using floating point notation, if I in-
crease the number of bits I use in my rep-
resentation, I increase the range of
rational numbers I can represent. But I
still cannot accurately represent any of
these "irrational" rationals. In fact, I will
encounter more of these "irrationals" in
my broader set. If, however, I represent
rational numbers as ordered pairs of in-
tegers, increasing the number of bits I use
increases both the range and accuracy.

The unit RATIONAL.PAS (see Figure

1) defines a rational number as an array
of integers and defines several operations
on rationals. I have not included error
checking for overflow problems because I
rarely use numbers large or small
enough to cause problems that frequent
simplification cannot remedy.

There are also no functions for testing
for a zero or negative in the denomina-
tor. Since multiplication is the only
operation performed on denominators,
these are not needed. The only exception
is in the INVERT procedure.

The program GAUSS.PAS (see Figure

2) uses both reals and rationals to do a
Gauss elimination on a system of two
equations in two unknowns. Run this
program yourself with different coeffi-
cients and notice the difference in results.
Try coefficients like 3, 6, and 21, as well
as primes.

Mr. Entsminger points out that no
system is capable of representing rational
numbers to an unlimited significance. No
matter which representation I use, I can
only represent a finite, discrete set of
numbers. Since one of the most signifi-

Continued on page 39

MICRO CORNUCOPIA, #53, May, 1990 35

Continued from page 35

else If (N mod 10) in [0,2,4,5,6,8] then

If X[l] = then begin

done := true;

X[l] := 0;

While not Done do

X[2] := 1;

If Divisor >= N then begin

end

Pr := true;

else begin

Done := true;

A := 6CD(X[1],X[2]);

end

X[l] := X[l] div A;

else if (( M mod Divisor) = 0) then

X[2] := X[2] div A;

Done := true

end;

else begin

End;

Count := (Count + 1) mod 4;

If count = 1 then

Procedure Invert ( Var X : Rational) ;

divisor := divisor + 4

Var

else divisor := divisor + 2;

A : longint;

end;

Begin

Prime := Pr;

If X[l] = then

End; {Prime}

WritelnC DIVIDE BY ZERO ERROR!')
else if X[l] < then begin

Function GCD( A, B : longint) : longint;

A := X[2];

Var

X[2] := Abs(X[l]);

x,M,N : longint;

X[l] := - A;

Done : boolean;

end

Begin

else begin

M : = abs (A) ;

A := X[l];

N : = abs (B) ;

X[l] := X[2];

x := M;

X[2] := A;

If N <= x then x := N;

end;

Done := false ;

End;

While not done do

If ((N mod X)=0) and ( (M mod X)=0) then

Procedure Radd( XI, X2: rational; Var Y: rational).

done :=t rue

Var

else x := x - 1;

Z,A : longint;

GCD := x;

Begin

End;

Z := LCM(X1[2],X2[2]) ;
Y[2] := Z;

Function LCM( A, B : longint ) : longint;

A := Z div XI [2];

Var

Y[l] := XI [1] * A;

X,Y,N,M : longint;

A := Z div X2[2];

Done : boolean;

Y[l] := Y[l] + ( X2[l] * A) ;

Begin

Reduce (Y) ;

M := abs (A) ;

End;

N := abs(B) ;

Y := M;

Procedure Rsub( XI, X2: rational; Var Y: rational),

If N < Y then Y := N;

Var

X := Y;

Z,A : longint ;

Done := false;

Begin

While not done do

Z := LCM(X1[2],X2[2]) ;

If ( (X mod M) = 0) and ( (X mod N) = 0) then

Y[2] := Z;

done := true

A := Z div XI [2];

else x := x + Y;

Y[l] := Xl[l] * A;

LCM := X;

A := Z div X2[2];

End;

Y[l] := Y[l] - (X2[l] * A);
Reduce (Y) ;
End;

Procedure Reduce ( Var X : rational) ;

Var

Procedure Rmult (XI , X2 : Rational; Var Y: Rational)

A : longint;

Var

Begin

A,B : Rational;

Continued on page

36 MICRO CORNUCOPIA, #53, May, 1990

DisiDoc v3.xx- REftLHODE.EXE

Help Load Forwat Edit
For«at: All P

Unlimited file

size. . . Fully

automatic

Batch mode and
interactive

■ow ex
■ov bx
;»»> Conversion Section
les di

Locates
Data/Code
boundaries

-aodtle-
Sabroatine
Hi in
Or if in

Xref address

Uord

Address

Patch Xoption | Dos Keep Quit_
<16> no patches

8662:6881 »xref=<86847>«
8682:8884 »xref=<86688>«

: conversion table
8682:8887
8682:8889
8682:888b

:Load register w 8
8682:888d
8682:6818

8682:8812 »xref=<86698>«
conversion table

8682:8816

8682:881b get byte count

8682:681f

;Store AL at EStCDU

9682:8821

; Turn OH Interrupts

8682:8822

8682:8825 D0S:4c-ter»iiwtte

Built-in BIOS
Preprocessor

All Data formats
including DB.DW,
DD&DUP

EXE Unpacker
included!

No Source Code? No Problem.
New DisOoc Professional is your dual-mode
key to any DOS source code. It works in batch
and interactive modes simultaneously, allowing
you to generate the core information of even the
most complex programs fast. . . and modify them
even faster. Most programs will come apart in just
two minutes. Imagine what you can do with a tool
this powerful! DisOoc sifts through programs
eight times for guaranteed accuracy. When code
gets mixed up with data, our toolbox comes to the
rescue with smart search and easy edit utilities.
DisOoc can handle any instruction set up to
and including 486 and offers a variety of other
great features that you can sample on our
Free Demo Disk.

Warning: DisOoc Professional may change the
way you work forever.

Programmers who used to shy away from fixing
outmoded programs with no source code are
going to discover a valuable new talent the ability
to modify and revise codes that would cost way

too much to start over (it's a programming man-
ager's dream). Save your employer huge new-
programming fees and enhance your marketabil-
ity. DisOoc is so easy to learn, you'll be a
high-dollar hero in no time!

Knowledge really is power.
Dis«Doc Professional is an amazing new teach-
ing tool. Learn how programs work. . . take them
apart and see the writing techniques that top pros
use. Use it to assist in debugging. Hunt down vir-
uses and write killers. DisOoc can save you
years of frustration, and it only costs $149.95
including the EXE Unpacker (until January 1). To
order your DisOoc Professional kit or our free
demo disk, simply call:

1-800-446-4656

WITHIN CT & OUTSIDE THE U.S., CALL (203) 953-0236

MasterCard & VISA • Shipped Immediately Via UPS

RJSwantek, Inc., 1 78 Brookside Road • Newington CT 061 1 1

Continued from page 36

Begin

{Subtract D*l from 2}

A[l] :> XI [1];

x := d;

A[2] := X2[2];

Rmult (x, a, y) ;

B[l] := X2[l];

Rsub(d,y, z) ;

B[2] := XI [2];

d := z;

Reduce (A) ;

Rmult (x, b, y) ;

Reduce (B) ;

Rsub(e,y, z) ;

Y[l] := A[l] * B[l];

e := z;

Y[2] := A[2] * B[2];

Rmult (x,c,y) ;

Reduce (Y) ;

Rsub(f ,y, z) ;

End;

f := z;
{Divide 2 by e}

Procedure Rdiv(Xl,X2: Rational; Var Y: Rational) ;

x := e;

Var

Rdiv(e,x,y) ;

X : Rational ;

e := y;

Begin

Rdiv(f,x,y) ;

X := X2;

f := y;

Invert (X) ;

{Subtract b*2 from 1}

Rmult (XI, X,Y);

x := b;

End;

Rmult (e,x,y) ;

End.

Rsub(b,y,z) ;
b := z;

♦ ♦ ♦

Rmult (f,x,y) ;
Rsub(c,y, z) ;
c := z;

Figure 2— GAUSS.PAS

Writeln ('Using rationals we get...');

Writeln(a[l],'/',a[2],'X = ',c[l],'/'

,c[2]);

Program GAUSS ;

Writeln(e[l],'/' ,e{2], 'Y = ',f[l],'/'

,f[2]);

Uses RATIONALS;

writeln;

Var

{Using reals} {Divide 1 by a}

A,B, C,D,E,F,X, Y, z : Rational;

xr := ar;

Ar,Br,Cr,Dr,Er,Fr,Xr,Yr, Zr : Real;

ar := ar / xr;

Begin

br := br / xr;

A[l] := 131;

cr := cr / xr;

A[2] := 1;

{Subtract D*l from 2}

Ar := 131;

xr := ar * dr;

B[l] := 23;

yr := br * dr;

B[2] := 1;

zr := cr * dr;

Br := 23;

dr := dr - xr;

C[l] := 31;

er := er - yr;

C[2] := 1;

fr := fr - zr;

Cr := 31;

{Divide 2 by e)

D[l] := 7;

xr := er;

D[2] := 1;

dr := dr / xr;

Dr := 7;

er := er / xr;

E[l] := 27;

fr := fr / xr;

E[2] := 1;

{Subtract B*2 from 1}

Er := 21;

xr := br * dr;

F[l] := 67;

yr := br * er;

F[2] := 1;

zr := br * fr;

Fr := 67;

ar := ar - xr;

{Using Rationals) {First divide 1 by a}

br := br - yr;

X := A;

cr := cr - zr;

Rdiv(a,x,y) ;

Writeln (' Using REALS we get');

a := y;

Writeln;Writeln(ar:5:9,' *X = ',cr:5

9);

Rdiv(b,x,y) ;

Writeln (er:5: 9, ' *Y = ',fr:5:9);

b := y;

End.

Rdiv(c,x,y) ;

c := y;

♦ ♦ ♦

38 MICRO CORNUCOPIA, #53, May, 1990

cant properties of the rational numbers is
that given any two rationals a and b,
there must be a rational z such that a< z<
b, I can never represent a complete range
of the rationals. In modeling natural sys-
tems, I do not need to.

The set of all numbers which exist in
nature is discrete. In making such a bold
claim, I feel compelled to defend it.

Using the most accurate measuring
instruments available, and the best tech-
niques of extrapolation, you can measure
a quantity to only finite tolerance. You
will then be able to get two measures for
which no intermediate measure can be
found.

Consider Zeno's paradox. Zeno
asserts that if a turtle and a rabbit were
to run a race, and the turtle were given a
head start, the rabbit could never pass
the turtle. His reasoning is that when the
rabbit achieved the turtle's starting posi-
tion, the turtle would have moved some
distance ahead. The distance between the
two, though less, would still be positive.
Continuing in this manner, Zeno says
that the turtle must always be some posi-
tive distance ahead.

This is true only if distance is in-
finitely divisible. Try this experiment
yourself with two objects moving at con-
stant velocity, one faster than the other.
Give the slow one a head start and you
will see that the fast one does pass the
slow one. The distance between them re-
duces to a point where no smaller dis-
tance exists.

Mr. Entsminger's example of the
equation Nextx + RX(1 - X) yields values
between and 1, representing some por-
tion of the maximum population of
wolves. If the maximum number of
wolves is some number z and the popu-
lation after any iteration is given by
z*Nextx, we see that values of Nextx of
minute enough significance are not use-
ful. Suppose Nextx = (K(z + 1) + l)/(z +
1) for some rational K. We then have a
fraction of a wolf, which is not likely to
reproduce.

For many applications, all discrete
representations of the rational numbers
are equally flawed. Trigonometric equa-
tions, for example, rely on the irrational
number pi. The results obtained using
ordered pairs will be at least as inaccu-
rate as those obtained using floating
point notation. But for many applica-
tions, representing rational numbers as
ordered pairs of integers will greatly in-
crease accuracy.

♦ ♦ ♦

CITIZEN PRINTERS

Citizen 120D 120/25 CPS 80 COL $165.00

Citizen 180D 180/35 CPS 80 COL $180.00

Citizen HSP500 300/66 CPS 80 $350.00

Citizen HSP550 300/66 CPS 120 $475.00

Citizen GSX140 200 CPS 24 PIN $345.00

GOLDSTAR MONITORS

MBM-1210A 12" TTL AMBER $ 90.00

MBM-1401A 14" Flar Screen TTL $135.00

MCH-1420 14" EGA 640x350 $345.00

MCH-1430 14" VGA 640x480 $375.00

VIDEO CARDS
ATI VIP Card VGA , CGA , EGA , MGA ,

ANALOG AND TTL 800x560 $175.00

ATI VGA Wonder 256 1024x768 $225.00

ATI VGA Wonder 256 16 bit $375.00

Paradise Basic EGA 640x350 $ 99.00

Paradise Basic VGA 640x480 $175.00

Paradise VGA Plus 800x600 $199.00

Paridise VGA Plus 16 Bit $275.00

HARD DRIVE CONTROLLERS DOR XT & AT

WD WX1 Auto Config PC/XT
WD WX2 Auto Config PC/XT
LONGSHINE LCS-6210D PC/XT
WD WAH AT 16 Bit 2 Hard Drive
WD 1003SM1 AT 2:1 Int.
WD 1003SM2 Hard/Floppy 2:1
WD 1006SM1 AT 1:1 8K Cashe
WD 1006MS2 Hard/Floppy 1:1
WD 1003SR1 RLL 2:1 Int
WD 1003SR2 RRL Hard/Floppy
WD 1006SR1 RLL 1:1 8K Cashe
WD 1003SR2 RLL Hard/Floppy

65.00
55.00
45.00
75.00
99.00
$125.00
$125.00
$135.00
$125.00
$135.00
$135.00
$145.00

SEAGATE DRIVES AND CONTROLLER IN STOCK
Please call for current price.

********** 2400 BAUD MODEM **************
by Computer Peripherals
2400 Baud Internal and Software $85.00
2400 Baud External and Software $140.00

CASCADE ELECTRONICS, INC.
ROUTE 1 BOX 8
RANDOLPH, MN 55065
507-645-7997

Please ADD shipping on all Orders

COD Add $3.00 Credit Card ADD 5%

MN Add 6% Sales Tax Subject to change

Reader Service Number 15

MICRO CORNUCOPIA, #53, May, 1990 39

Initializing Variables In Turbo Pascal

Handling Initialization During The Link Rather Than At Run Time

This is hacking at its best. Dave ob-
viously understands the limitations of Turbo
and he's sent along some fixes. Fun stuff.

The most recent versions of Turbo
Pascal (5.0 and 5.5), have a great
deal to offer in terms of flexibility,
code readability, and development
speed. However, because of the way the
compiler and its internal linker work,
and because of Borland's support for
only one memory model, it has some
real limitations.

Procedure Tables To Go

One of Turbo Pascal's annoying limi-
tations is the compiler's inability to ini-
tialize arrays of procedures (procedure
tables) at compile time.

Procedure tables are easy in C. The ex-
ternal linker automatically initializes the
address values in an initialized array of
procedures (at link time after they have
been resolved). See Figure 1 for a simple
example. (I'm not a C programmer by
trade, so the example may not be ele-
gant.)

If you were to translate this directly
into Pascal (see Figure 2) and then try to
compile it, Turbo complains about the
line that initializes the array of proce-
dures.

So what do you do if you want to cre-
ate a procedure table in Turbo Pascal?
Normally you have to waste some run-
time code to assign each of the proce-
dures to the array members (see Figure
3).

For an array of four procedures, this
doesn't add much to the run-time. But
how about 500 procedures? Then it adds
up. The .EXE file for the small interpreter
you just knocked together using proce-
dure tables would be fatter than it should
be. Plus, I think if s kludgy to do all those

Figure 1— PROCTABL.C

void procO (void)

puts ("procedure 0") ;

void procl (void)

puts ("procedure 1");

void proc2 (void)

puts ("procedure 2");

?:»3-; f :: : i;v :! ;

void proc3(void)

■ i ':.-.

puts ("procedure 3");

1/ ■]■-:.'

void static (*proc[] ) (void) = {procO, procl, proc2, proc3 } ;
main()

int n;

for(n=0;n<4; (proc[n]) (),n++);

Figure 2 — Pascal Version of Figure 1

PROCTBL0 . PAS (26) : Error 99: File and procedure types not allowed here.
Proc : array[0..3] of procedure = (ProcO, Procl, Proc2, Proc3) ;

{$F+>

procedure ProcO;

begin

writeln (' Procedure 0' ) ;
end;

procedure Procl ;
begin

writeln (' Procedure 1' ) ;
end;

procedure Proc2;
begin

writeln ('Procedure 2');
end;

procedure Proc3;

40 MICRO CORNUCOPIA, #53, May, 1990

By Dave Gwillim

159 Woodbury Road

Hicksville, NY 11801-3030

(516) 942-8697

begin

writeln ( ' Procedure

3');

end;

<$F->

const

Proc : array [0.. 3]

of procedure

= (ProcO, Procl, Proc2, Proc3) ;

var

n : word;

begin

for n := to 3 do

Proc [n] ;

end.

♦ ♦ ♦

Figure 3— PROCTABL1.PAS

var

Proc : arrayfO. .3]

of procedure;

n : word;

{$F+}

procedure ProcO;

begin

writeln ( ' Procedure

0');

end;

procedure Procl;

begin

writeln (' Procedure

i');

end;

procedure Proc2;

begin

writeln (' Procedure

2');

end;

procedure Proc3;

begin

writeln ( ' Procedure

3');

end;

{$F-}

begin

Proc[0] :«« ProcO;

Proc[l] := Procl;

Proc [2 J :« Proc2;

Proc [3] :- Proc3;

for n :» to 3 do

Proctn];

end.

♦

♦ ♦

ne of Turbo
Pascal's annoying
limitations is the
compiler's inability
to initialize arrays
of procedures
(procedure tables)
at compile time.

assignments, especially when the linker
knows what the addresses are before it
creates the .EXE file.

Is there a way to make the compiler
handle this at compile time? Using only
Turbo Pascal, no. The compiler handles
the initialization of typed constants
before linking. At that time it isn't sure
where the procedures will be located.

However, by using a dash of as-
sembly language, you can force the com-
piler to handle most of the initialization
during the link, leaving only the initiali-
zation of a single pointer variable for the
run-time code. Figures 4 and 5 show one
way you can do this.

The fake procedure Procs is never
called (and never should be, unless you
feel like reaching for the red button). If s
simply used to export a pointer from an
external assembly language program.
Since you cannot declare any data PUB-
LIC in an external procedure and still
have Turbo Pascal see it, you have to
cheat.

Fortunately, cheating will make your
.EXE file smaller, especially when you

MICRO CORNUCOPIA, #53, May, 1990 41

have lots of procedures in your table. The
trade off is that you have to maintain a
PROCS.ASM file in addition to the Pascal
source code.

Typed Constants Without Limit,
Anyone?

We all know how valuable that 64K of
data segment for variables in Turbo Pas-
cal is, right? Every typed constant (initial-
ized variable) gets its bite (or two). Some-
times 64K seems downright tiny!

Of course, you can allocate all your
variables on the heap. But then you have
to initialize them with assignments to
constants embedded in your code or load
in a separate data file. Both approaches
add considerably to the size of your .EXE
file.

You don't have this problem in any
implementation of C that supports multi-
ple data segments. So wouldn't it be nice
if Turbo Pascal could manage it? If only
you had some way to assign initialized
variables to their own segment. Perhaps
you wouldn't have to switch to C!

Using a little assembly language, you
can do this in Turbo Pascal. Using only 4
bytes of that valuable Turbo Pascal varia-
ble space you get access to as much as
64K of initialized data. This trick turns
code space into initialized data space. See
Figures 6 and 7 for a way to do this.

Note that to get 64K of initialized data
space using this technique, you would
have to make the external procedure con-
taining the data the sole member of a
separately compiled unit. And do it so
that virtually none of the code space
would be taken up by executable code.

When unitizing, you should also in-
clude the pointer variable in the unit and
use the run-time initialization capability
(placing an assignment statement be-
tween a BEGIN and END in the im-
plementation part) to set up the pointer
variable. You only use the unit to have all
the initialized data available. Using the
unit approach is also a good way to hide
the fake procedure _DataSegl .

See Figures 8 and 9 for a sample using
the unitized approach.

Putting the initialized data into the
.ASM file also means you can use the
power of TASM or MASM operators and
macros to generate complex patterns of
initialized data that Turbo Pascal can't
match. Just try initializing a typed con-
stant with the equivalent of 1000
dupCTEXT ') in Turbo!

Figure 4— PROCS.ASM

public Procs

; import the addresses for the Pascal procedures
extrn ProcO : far, Procl : f ar , Proc2 : far, Proc3 : far

code segment
assume cs : code

Procs proc far

; use the addresses to build a procedure table
dd ProcO
dd Procl
dd Proc2
dd Proc3

Procs endp

code ends
end

Figure 5— PROCTBL2.PAS

{$F+>

{$L PROCS. OB J}

procedure Procs ; external ;

type

ProcArray « array [0.. 3] of procedure ;

Proc : ^ProcArray;
n : word;

procedure ProcO ;
begin

writeln(' Procedure 0');
end;

procedure Procl ;
begin

writelnC Procedure 1');
end;

procedure Proc2;
begin

writelnC Procedure 2');
end;

procedure Proc3;
begin

writeln { ' Procedure 3 ' ) ;
end;

begin

Proc := @Procs; { point Proc to the procedure table in PROCS. OB J }

for n :<= to 3 do
Proc* £n] ;
end.

♦ ♦ ♦

Figure 6— DATASEG1.ASM

public _DataSegl

code segment
assume cs : code

_DataSegl proc far

DefaultlnputFile db 11, 'DATAJtN.DTA' , 68 dup(0) ; string£79]

42 MICRO CORNUCOPIA, #53, May, 1990

DefaultOutputFile db 12, 'DATA OOT.DTA', 67
InitFileOffset dd 1623447 ~
BigArray db 1000 dup ('TEXT ')

dup(0) ; string[79]
; longint
; array [1. .5000] of char;

_DataSegl endp

code ends
end

♦ ♦ ♦

Figure 7— INITDATA.PAS

{$F+)

{$L DATASEGl.OBJ)

procedure DataSegl; external;

t$F-}

type

DataSeglType = record

DefaultlnputFile : string [79] ;

DefaultOutputFile : string [79];

InitFileOffset : longint ;

BigArray : array [1. .5000] of char;
end;

{ 'DATA_IN.DTA' }
{ 'DATA OUT.DTA' }
{ 1623447 }
{ 1000 dup ('TEXT ') }

var

DataSegl : A DataSeglType;
n : word;

begin

DataSegl := @_DataSegl; {point to initialized data in DATASEGl.OBJ}

with DataSegl* do

begin

write In (Default InputFile) ;
writeln (DefaultOutputFile) ;
writeln (InitFileOffset) ;
for n := 1 to 30 do

write (BigArray [n] ) ;
writeln;
end;
end.

♦ ♦ ♦

Figure 8— DATA1.PAS

unit Datal;

(*************) interface (

**************

type

DataSeglType - record

Default InputFile : string [79];

DefaultOutputFile : string[79];

InitFileOffset : longint;

BigArray : array [1. .5000] of char;
end;

{ 'DATA_IN.DTA' }
{ 'DATA OOT.DTA' }
{ 1623447 }
{ 1000 dup ('TEXT ') }

var

DataSegl : "DataSeglType;

(*************) implementation

(ft************)

{$F+>

{$L DATASEGl.OBJ}

procedure DataSegl; external;

{$F-}

begin

DataSegl := @_DataSegl; {point to initialized data in DATASEGl.OBJ}
end.

♦ ♦ ♦

Confusing Code?

(Mnclude <stdio.h>

text count(){int c.nlines.nwords, nchars, inword;

inword-NO;nlines-nwords-nchars-0;whtle((c

-getchar())l-E0F){^nchars;1f (c-'Vn'j

♦♦nltnesjtf ((c--' ')! !(c«'\n'))inword-N0;else

lf(inword--NO){inword.YES:++nwords;}}prlntf(

"*d Vd %d\n", nlines, nwords, nchars);)

C It Your Way!

#inc

lude <stdto.h>

text

_count ()

Int c, nlines, nwords.

nchars.

i nword

inword - NO;

nlines - nwords - nchars - 0;
while ((c - getchar()) 1- EOF
♦♦nchars;
if (c -- '\n')
♦♦nlines;
if ((c «'•)!! (c -•

inword - NO;
else if (Inword — NO) (
inword - YES;
♦♦nwords;
)

{
\n'))

)

}

printf ("W W M\n", r

lines,

nwords ,

nchars);

MD\A/ with c " c,ear| y™' f o rm at C source

II UW code exactly the way you want it.
C-Clearly's context sensitive analysis will
format any C program in your own personal
or corporate style.

F AQV t0 use ' C" c,ear| y' s st y' e templates
EZMO I are a snap to modify, since they
resemble C source code you edit into your
preferred format. Templates are included for
several common styles as well as standard
K&R.

I |QT||J/^Q can also be created with
LIO I IflUO function names and
comments highlighted for improved readability.
Listing options include line numbers, headers
and/or footers and flow lines.

I fl C A I for makin Q obtuse code clear.
IUCML Allows all of your source code
to be presented in a consistent format of your
choosing. Also great for code walkthroughs and
final documentation listings.

lAfHDIf Q with a " ,BM PC - ^ AT - ps/2

WUnrVO and compatibles, with 51 2K
RAM. Automatically processes all include files
and pre-processor statements. ANSI-C com-
patible. Not copy protected.

C-Cleatly $129.*

Shipping & Handling USA S5; Canada/Mexico $10; Other Countries
$15; CA Residents add sales tax. Visa/MasterCard/COD accepted.

For orders or information call:

1-800-662-8266

V COMMUNICATIONS, INC

4320 Stevens Creek Blvd., Ste 275, Dept. MC6
San Jose, CA 95129 (408) 296-4224

Reader Service Number 62
MICRO CORNUCOPIA, #53, May, 1990 43

Initializing Large Amounts Of Data

For large amounts of initialized data,
a program that reads the data structure
definition and then creates the appro-
priate db, dw, dd directives, etc., would
make creating the .ASM file easier. You
might write an include file that has the
initialized data appear after each of the
members in the record structure defini-
tion. Then feed this include file to your
program as a data file and have it create
the .ASM file.

See Figure 10 for an example.

Your program could parse each line to
determine the equivalent assembly lan-
guage data type and number of elements
and then initialize it with the data in the
comments. As the famous cop-out goes,
"Such a program is left as an exercise for
the student."

Editor's note: Yes, yes. I enjoy writing in
Turbo Pascal but I, too, am running up
against limitations in the language, especially
with the 64K limit on the data segment.

♦ ♦ ♦

Figure 9— USEDATA.PAS

uses

Datal;

SSviS^SSSS^S^SS^^^S^^S^^^^SS^^^^^^^^^K^^R^^P^^^^^^S

n : word;

begin

with DataSegl A do

begin

writeln (Default InputFile) ;

writeln (Default OutputFile) ;

writeln (InitFileOff set) ;

for n := 1 to 30 do

write (BigArray [n J ) ;

writeln ;

: ^S§SM^MSX8'^^^^^^B§SiSS'^S^^X^WS§^M^^^§^SiSMW^^S^X:

end.

; ;^#SC#®

Figure 10 — Example Include File Layout

DataSeglType = record

DefaultlnputFile : string [7 9]; {= "DATAJEN.DTA" }

DefaultOutputFile : string£79] ; {= "DATA_OUT.DTA" }

InitFileOffset : longint; {= 1623447 }

end;

^^^^^M^M^^^^^S^^^MMf^MM^EMiM^Z^MM^Mi^M^^^^MMiM

Multipurpose Lab Interface for Your /SAP

Turn your IBM-compatible computer into a powerful laboratory
instrument with our Multipurpose Lab Interface (MPLI) hardware
and software. MPLI allows you to use a wide variety of sensors to
make measurements and to analyze and graph the results. Data
can be collected at a rate of thousands of samples per second or
slowly over periods as long as several months. Sensors available
include assembled, ready-tc-use units and parts kits. The MPLI
hardware-software combination consists of:

A 12-bit A/D Interface Board: A 12-bit, 8-input analog-to-digital

converter with built-in sample and hold. Conversion time of the A-to-D is 9 usee. There are three bipolar
and three unipolar software-selectable voltage ranges. The board also includes one 1 2-bit analog output
and access to 16 digital I/O lines. Includes manual and sample programs disk. (Order Code AIB-PC, $240)

Multipurpose Lab Interface Box: Includes three 8-pin DIN sockets which allow quick connection to 3 of the analog inputs, voltage out and
power leads. A prototyping area for building your own circuits is included on the circuit board inside the box. A variety of sensors and probes
are available which plug into the MPLI box: pH, thermocouple, temperature (AD590), force, microphone, etc. (Order Code BOX, $45)

Multipurpose Lab Interface Software: The Multipurpose Lab Interface Program allows three input
channels to be calibrated to display any type of input signal; for example, channel A can read temperature,
channel B can read pH , and channel C can read pressure. Data from each of the channels can be collected,
graphed, and saved on disk. The output voltage can be controlled by the program. The Oscilloscope mode
lets the computer act as a triple-trace, storage oscilloscope with a sampling rate of at least 40,000 samples/
sec (on an 4.77 MHz. IBM PC), much faster on other computers. The program even does X vs. Y plotting,
so you can display Lissajous figures on the monitor. (Order Code MPP-IBM, $49.95)

The IBM MPU Package price of $310 Includes:

• 12-bit Interface Board with manual and sample programs

• MPLI Interface Box, cable and test leads

• MPLI Program with extensive manual

A similar package is available for the Apple® II computer.

Call, write or fax for more information.

Vernier Software

2920 S.W. 89th Street #C

Portland, Oregon 97225

(503) 297-5317 FAX (503) 297-1760

Reader Service Number 208

44 MICRO CORNUCOPIA, #53, May, 1990

ERAC CO.

P.O. Box 1108, 14179 Halper Road
Poway, California 92064 ■ (619) 679-8360

Baby 386-20/24
Motherboard

One 32 bit-slot, Five 16-bit slot Two 8-bit slot Maximum
8 meg on board (1 meg SIMMS). OK on board. 80287 or
80387 or Weitec Co-Processor slot Shadow RAM.

-20 $630

-24 $650

1 meg SIMMS-80ns $115

16 BIT VGA Card

RESOLUTION COLOR

640x480 256 of 256*

800x600 16of256*

1024x768 16of256*

•with 51 2K RAM

Runs all standard VGA modes, 256K RAM on board.
Drivers for AutoCad, Lotus, Framework, GEM, VP, WP,
WS & MS Windows. 1 year warranty.

For 51 2K $30 extra $185

MOTHERBOARDS 386DX/SX/286

DX5000386/25MHz-32KCache-16 Meg-Simms-AMI Bios-Full $996

DX4000 386/20/24MHZ Shadow-8 Meg-Simms-AMI-Baby 630/650

P9400 386SX/16MHZ 8 Meg-Simms-Phoenix-Bios-Baby 375

P9200386SX/16MHz2Meg-Dip-AMI-Bios-Baby 365

SUPABOARD286/12MHz4 Meg-Dip-DallasClk-AMI-Baby 197

I/O, KB, CONTROLLERS, Etc.

XT Enhanced Hard Disk Drive Controller with Cables 48

DC-11M AT Hard/Floppy Contr. 1:1 438Kb/Sec,To2048Cyl 93

AT I/O PLUS 1 Par(Lpt 1 -3), 2 Ser(Com 1 -4), Game, Cables, Ser 2(opt) 38

VGA-16 16-brt VGA Board 800x600 with Driver Software 118

2400 Baud External Modem, with Software and Manual 99

KB5161 AT/XT 101 Keyboard, Cherry Keyswitehes (Click) 47

COLORMOUSE Black-Red-Blue-Beige-Green-Yellow, Software & Manual 39

VOICE MASTER KEY, Add voice commands to software XT/AT 1 47

MONITORS

VGA 1489 1024x768, .28 Dot 14", Swivel,

Hor.31/35KHz, Vert 50, 60, 70, 87 Hz 430

CASES

MINI-TOWER, 230 W, Reset, Turbo, Keylock, Speaker 143

TOWER, 230W, Reset, Turbo (2 Dgt), Keylock, Speaker 229

KAYPRO Equipment
Bargains

9" Green Montor-83, 84, K16 $60

Hostlnterface Board 15

Keyboard 50

Replacement Power Supply 70

Drivetek 2.6M FDD (Robie or K4X) 75

We Repair CPMKaypros

CPM COMPUTERS

K4-84 425

K10 495

K4X 425

We carry all IC's for Kaypro
repair.

IC's

81 -189 Video Pal $15

81-194RAMPal 15

81-Series Char. Gen. ROMs 10

81 -Series Monitor ROMs 10

Test Equipment

OSCILLOSCOPES

TEK7403N/7A18N/7B50A60 MHz . . .$650
Leader LB0520 30 meg Dual Trace .... 300

TEK 475 Dual Trace 200 MHz 1250

Scope Probe x1, x10 100 MHz 25

ANALYZERS

TEK 491 10MHz-40GHz $3500

HP851B/8551B10HMz-40GHz 1500

Biomation 805 Waveform Rcrdr 195

Biomation 81 00 2-Channel

Waveform Recorder 295

HP1600A Logic Analyzer 16ch 295

HP1600A/1607A Logic Analyzer 32ch . . 495
Gould K40 32ch Logic Analyzer 750

MISCELLANEOUS

Optronics 550 MHz Freq Cntr $95

Heatgun 120Vac7 A 35

TERMINALS

Televideo925 $99.00

HOURS: Mon-Fri 9-6. Sat 10-4
For more information, please call

NiCds

AA Cells .6ah $1.00

12V Pack AA Cells Jah 6.50

Sub-C Cells 1.5ah 1.50

12VPackSub-C 10.00

Double D Cell 2.5V 4ah unused 8.00

CCells 1.75

7.2V RC-Pack 1 .2ah 18.00

GEL CELLS

6V5ah 5.00

6V8ah .'...6.00

12V15ah 15.00

12V2.5ah 8.50

DCell2.5ah 2.00

ROBOTICS

5V DC Gear Motor w/Tach 1"x2" $7.50

Z80 Controller with 8-bit A/D 15.00

12V Gear Motor 30 RPM 7.50

Cable: DB9M-DB9F V length 2.00

High Voltage Power Supply

Input 15-30V DC

Output 100V400V16KV 6.50

SWITCHERS

AT 200W Pulls, tested $35.00

5V/75, 12V/6, -12V/3, -5V/5 85.00

5V/9.5A, 12V/3.8A, -12V/.8A 39.00

5V/3A, 12V/2A, -12V/.4A 19.50

5V/6A, 12V/2A, -12V/1A 29.00

5V/6A, 24V/1 1/4A, 12V/.6A, -12V/.6A . . 29.00

5V/30A 39.00

5V/100A 100.00

5V/120A 110.00

HP DC/DC 12Vin,5V/8A,12V/5A,-5V/.3A . 45.00

VERSATEC8222F22"

Electrostatic Printer Plotter

200 dots per inch. Up to D size.

1"per second $2,999

AT 80286-6 CPU BOARD

with reset and mono/color switch. Connector for
KB, Battery & SPKR. Phoenix Bios (tested with
Award 3.03), 6MHz, can be upgraded to 8 or
10MHz. Use with backplane, add memory
board, I/O boaid, etc.
ONLY $99

(619)679-8360

MINIMUM ORDER $25.00

TERMS: VISA OR MasterCard (Add 3%), Certified Checks, Money Order, NO COD.Personal Checks

must clear BEFORE we ship. Include shipping charges. California residents add 7 1/4% Sales Tax.

Reader Service Number 93

MICRO CORNUCOPIA, #53, May, 1990 45

Fast Neural Networks

Transputers & Code Optimization Make Them Almost Cheap

You can run neural networks on 286s
and 386s; but as the networks grow, the time
required for training and running gets
significant. Fortunately neural nets lend
themselves to parallel processing, and
fortunately there are transputer boards out
there that don't cost megabucks.

Although not the ultimate solution,
a coprocessor board (CPB)
plugged into a PC (or Macintosh)
can significantly increase the speed of a
neural network for a relatively low cost.

The CPBs I'll discuss and program in
this article are characterized by one or
more high speed digital signal proces-
sors and a large amount of memory.

A few companies market CPBs (the
SAIC Delta II, the HNC Anza Plus, etc.)
specifically for neural nets. These CPBs
come with software for implementing
several different neural networks. One
of these systems costs $10,000 plus.

Other CPBs (Mercury cards, transput-
ers, etc.) are marketed for general
numerical and digital signal processing.
They're a little less jazzy, but cheaper.

Transputers, in particular, are useful
for optimizing neural networks because
they're designed for parallel processing.
The transputer has the advantage of low
start up cost ($3,000 for a single trans-
puter with 1M memory, board, and com-
piler). However, its performance isn't as
good as some of the more expensive sys-
tems. You can overcome this deficit,
however, by ganging transputers.

In this article I'll discuss the trans-
puter in general, then show you how to
program individual and parallel trans-
puters. In particular, I'll use every trick
in the book to optimize the matrix-vector
multiply operation (the key neural net-
work process). Finally, I'll discuss a
group of programs which create a neural

network — a neural net that can run on as
many transputers as you want.

The Transputer

Inmos Ltd. makes several variations
on their transputer. The T800 model, the
one I'll discuss here, has both an internal
32 bit integer processor and an internal
32 bit floating point unit (see Figure 1).

The 20 MHz version of the T800 can
compute at rates up to 1.5 million single
precision (32 bit) floating point opera-
tions per second (faster than either the
68020/68881 or the 80386/80387).

This transputer has 4K of internal
RAM, an external 32 bit address space,
and 32 bit wide data. It contains special
communication hardware (serial DMA
channels called links) that allows many

transputers to work in parallel.

This transputer also has a couple of
timers as well as hardware supported
multitasking. The transputer was de-
finitely designed to run numerically in-
tensive applications.

Inmos-SGS Thomson and several
other vendors have developed products
that incorporate transputers into PCs.
Each transputer comes on a small
daughter board called a TRAM.

So far TRAMs exist for the PC, Mac II,
VME, and Eurocard buses. Depending
on memory configuration, up to ten
TRAMs will fit onto a PC motherboard.

Interprocess Communications

You don't have to decide if two
processes will reside in the same proces-

Figure 1 — Block Diagram of the T800 Transputer

ANSI-IEEE 754-1 985
Floating Point Unit

32 bit Integer Processor

20 Mbits/s

Bidirectional Serial
DMA (Link)

Control Logic
and Timers

4 K Bytes
Internal Ram

Bidirectional Serial
DMA (Link)

20 Mbits/s

External Interface

Multiplexed Address and Data bus

46 MICRO CORNUCOPIA, #53, May, 1990

By D. Gilbert Lee

Johns Hopkins University

Applied Physics Laboratory

Laurel, MD 20707

sor when you write the code for the
processes; the code is the same.

Processes communicate through one-
way channels. In theory, the only way
for a process to access data that origi-
nates from another process is through a
channel (but you can get around this if
you want to).

For a channel between processes on
adjacent transputers, we implement the
channel using a point-to-point hardware
link. Each link can either transmit or re-
ceive a serial stream of data at 20
Mbits/sec. There are four pairs of links
(half of each pair for transmitting, the
other for receiving) on the T800. We can
therefore attach a processor to a maxi-
mum of four other processors.

While individual links are capable of
only 1.7 Mbytes/sec throughput (2.4
Mbytes /sec when you use both links
simultaneously), the aggregate com-
munications bandwidth of an array of
transputers is phenomenal. For example,
if you connect all the links of 16 trans-
puters, you can move over 4.8 gigabytes
of data per second.

You can connect transputers in
various arrangements (e.g., pipes, rings,
meshes, trees, hypercubes, etc.). This
differs significantly from multiproces-
sors with shared communications buses
that can handle only a few additional
processors before bus contention bogs
things down.

To create a channel between two
processes within the same processor, use
a memory location to hold the flag for
the microcode that implements the com-
munications instructions.

Communications are always synchro-
nized. When one process wants to com-
municate with another, it must wait
until the other is ready. If there are two
processes (whether local or across adja-
cent transputers) and one is waiting to
communicate, it will be suspended.

ransputers, in
particular, are
useful for optimizing
neural networks
because they're
designed for
parallel processing.

When the second process is ready to
communicate, data transmission begins.
The first process can then continue. It
doesn't matter whether the first process
was waiting to transmit or receive.

Multitasking

A microcoded scheduler manages
multitasking. The transputer supports
two levels of process priority: high and
low. Processes are either active or inac-
tive. The scheduler maintains two lists of
the active processes: one for high prior-
ity and one for low.

The scheduler runs all active high pri-
ority processes one at a time until
they're all waiting for communications
or a time delay. Then the process at the
top of the low priority list executes for
one time slice (approximately one milli-
second), or until it becomes inactive as a
result of waiting for communication or
waiting for a programmed time delay.

Inactive processes don't require at-
tention (i.e., time) from the scheduler.
After a communication has completed or
the programmed time delay has tran-
spired, the scheduler will add the

process back to the list of those waiting
to execute.

In sum, the use of the internal chan-
nels and the multitasking scheduler al-
lows a single transputer to support
several concurrently running processes.
The use of the external channels extends
the concurrency across multiple trans-
puters.

Programming Languages

Several languages support the trans-
puter either individually or in parallel
configurations: Occam, the first lan-
guage developed for the transputer (it
supports strong data typing, multitask-
ing, multiprocessing, and intertask I/O);
FORTRAN; C; Pascal; and Ada.

Each of these compilers extends its
language or provides libraries of
routines to support the transputer. The
transputer programming environment is
similar to any PC programming environ-
ment.

The C compiler I use, developed by
Logical Systems Inc., supports com-
munications and multitasking. It gener-
ates assembly language output for the
assembler. This means that I can read
the code the compiler generates. I can
then use the inline assembly language
support of the C compiler to optimize
the parts that need it. (Occam also sup-
ports "human readable" assembly lan-
guage.)

Optimizing Matrix-vector Multiply

You can improve neural network effi-
ciency by optimizing the matrix-vector
multiplies.

Most programmers use indices for
their matrix-vector operations to get
around the 80x86's 64K segments (see
Figure 2). The transputer can access data
anywhere within its 32 bit address space
and can have very large arrays. This sets
the stage for dispensing with indices and

MICRO CORNUCOPIA, #53, May, 1990 47

accessing the data via pointer arithmetic.

To use pointer arithmetic, you must
organize the data for the weight matrix
(in memory) in a very specific manner.

In practical terms this means you ar-
range all the weights for each next-layer
node consecutively. Then instead of cal-
culating an offset into the weight matrix
using indices for each weight, you incre-
ment a pointer to access each weight
sequentially (see Figure 3).

Using this approach with very small
arrays on an 8 MHz IBM PC/AT with
80287, 1 achieved 9800 multiply-accumu-
lates/sec (MAC/sec). On a 20 MHz
Compaq 80386 with 80387, I got 65,000
MAC/sec. This is about an 8% improve-
ment over an indexed system.

Next I experimented with the 20
MHz T800 transputer. With indices —
184,800 MAC/sec. With pointers—
276,900 MAC/sec. A significant
improvement, but mediocre considering
the advertised performance of 1.5
Mflops. Since a single multiply-accumu-
late operation executes two floating
point operations, the transputer should
be getting around 750,000 MAC/sec.
What's going on?

Because I perform the matrix-vector
multiply with floating point numbers,
it's crucial to keep the floating point pro-
cessor running continuously. The trans-
puter's architecture is designed such that
the integer processor can calculate the
addresses for the next set of operands
while the floating point processor ex-
ecutes a multiply or an add.

To do this you interleave the integer
processor instructions with the floating
point processor instructions. Since my C
compiler generates an assembly lan-
guage source file, I can see how effi-
ciently it uses the two processors.

Using inline assembly language, I can
replace the original C instructions with
assembly language instructions. In prac-
tice this turns out to be only six or seven
assembly language statements for a mul-
tiply-accumulate instruction.

You can also speed things up by
using two of the transputer's features:
how it creates constants, and its high
speed internal RAM. 3

Loops

Since so few operations occur per it-
eration during the matrix-vector mul-
tiply loop, a lot of time goes into
incrementing the index and testing
when to exit the loop. To save time you
can "open the loop" by enlarging the

48 MICRO CORNUCOPIA, #53, May, 1990

Figure 2 — C code fragment showing straightforward implementation of
Matrix- Vector Multiply.

IN[n] values for input nodes

W[m] [n] values of weights

OUT[m] values for next layer nodes

/* Initialize sum */

for(i =0; i < m; i++)
{

OUT[i] = 0;
for(j =0; j < n; j++)
{

OUT[i] = OUT[i] + (IN[j] * W{i][j]); /* Multiply and Accumulate */
}
}

♦ ♦ ♦

Figure 3 — C code fragment showing implementation of Matrix-Vector
Multiply using pointers only.

IN pointer to the n values for the input nodes

W pointer to the m x n values of weights

OUT pointer to the m values for next layer nodes

for(i =0; i < m; i++)
{

*OUT =0; /* Initialize sum */

IN = address of first input node
for(j =0; j < n; j++)

{

•OUT += *IN++ * *W++) ; /* Multiply and Accumulate */

}
OUT++; /* Move to next node */

}

♦ ♦ ♦

Figure 4 — C Code Fragment showing how the Multiply Accumlate loop
can be opened up (in this example by 4). The assembly language code
increments I and W on the fly instead of adding 4 at the end of the loop.

IN
W

pointer to the n values for the input nodes
pointer to the m x n values of weights

OUT pointer to the m values for next layer nodes

nr = n - (n%4) ;

for(i = 0; i < m; i++)
{

*OUT = 0;

IN = address of first input node
for(j =0; j < n%4; j++)
{
*OUT += (*IN * *W)

+ (*(IN+1) * *(W+1))
+ (*(IN+2) * *(W+2))
+ (*(IN+3) * *(W+3))
IN += 4;
W += 4;
}
for(j=0; j < nr; j++)
{

*OUT += *IN++ * *W++;
}
>

/* Determine remainder */

/
/* Initialize Sum */

/* Multiply and Accumulate */
/* Multiply and Accumulate */
/* Multiply and Accumulate */
/* Multiply and Accumulate */

/* Use normal approach */

/* for the last few */

/* Multiply and Accumulates */

code within the loop to perform several
multiply /add calculations (see Figure 4).

This approach will improve perform-
ance (to varying degrees) no matter
which processor you use. In fact, the
fastest code (assuming we're using a
non-caching processor) would require
no looping. Of course this is unrealistic
for all but the smallest neural network
because of the memory required for the
instructions.

How much should you open the
loop? You can determine this by taking
advantage of how the transputer works
with constants. The fastest executing in-
structions are a byte long and use the
first nybble to indicate the instruction
and the second nybble for a constant.
Thus, constants between and 15 are
easily handled in one-byte instructions.

If it needs larger numbers, the trans-
puter builds them a nybble at a time
(also using a byte-long instruction). The
instruction associated with the last
nybble necessary to form a constant will
also operate on the constant (e.g., load,
store, etc.). For example, if you need a
number between 256 and 4095, you need
three nybbles (with an additional nybble
instruction for each, i.e., a total of 3
bytes). Of course each additional nybble
requires an additional clock cycle.

Therefore a good amount to open the
loop up by is 16 (for constants between
0-15). This requires only one nybble (and
therefore one instruction) to access each
operand. To make the code usable for a
neural network of any size, add regular
loop code after the opened loop code to
handle the situation when the number of
nodes is not divisible by 16.

RAM

The second key feature of the trans-
puter is its internal RAM. Most com-
pilers use this RAM for stacks and
workspace pointers, but we can tell the
compiler to set aside space within inter-
nal RAM for a small routine or two. In-
structions and data found in this area
(4K) can be retrieved in a single clock
cycle.

Since the less expensive TRAMs use
external memory requiring four clock
cycles for access, we can speed up the
matrix vector multiply by using the in-
ternal RAM. It's too bad we can't get the
arrays in there also.

Now what are we up to? Using the
hand-optimized assembly language to
maximize the floating point operation
and opening the loop by 16 to minimize

So you want Tele to do it ALL...

OK, piece a cake!

Berry Computers presents The Tele Operating System —
A MULTITASKING MS-DOS COMPATIBLE

Computer software exists in layers.
Tele contains the layers between
your application and the hardware

Tele executes binary programs
intended for MS-DOS in a multitask-
ing, windowing environment.

You can use Tele to run several pro-
grams at once thereby eliminating
boring and unproductive delays.

For instance, in a software develop-
ment environment, you can set a
compiler running and simultaneously
edit another source file.

Tele uses a preemptive task scheduler.
Its file system is fully compatible
with MS-DOS. It runs on any pro-
cessor compatible with the Intel
8086 family.

Most programs intended for MS-
DOS will multi-task under Tele.

Microsoft could find no way to make
MS-DOS multi-tasking and still sup-
port all its existing applications.

Tele simply does the best possible.

Tele services the documented inter-
face to MS-DOS, fits in less than
100K of memory, and provides the
important features of OS/2., includ-
ing installable file systems.

Tele improves display performance
by a factor of 2, whether you use
multiple windows or not! On pro-
cessors slower than 20 MHz the
improvement is even greater.

Source code in C and assembly is
included for the interface to MS-
DOS. If you have to have MS-DOS
idiosyncracies, you can easily put
them in yourself.

Treat yourself to an operating
system where the marketing and
financial folks had no say at all in
the design. The result is an efficient,
structured program that either pro-
vides or specifically anticipates all
features of modern mainframe and
desktop operating systems.

Tele Operating System $100

Tele Tool Kit

Multitasking (MT)* $ 50

Windows (WI)* $ 40

File System (FS) $ 40

Tele Operating System
and Tool Kit $200

Demonstration Disk $ 5

*MT formally called SK. WI formally called CD.

Telephone support is freely available.

Tele is available from:

Crosby Associates

P.O. Box 248

Sutter Creek, California

95642

CALL NOW TO ORDER:
(209) 267-0362
FAX (209) 267-9246*

*Note new Fax #

Visa. Mastercard. American hxpress & Discover Card accepted.

MS-DOS and OS/2 are trademarks of Microsoft Corporation.

Reader Service Number 147

MICRO CORNUCOPIA, #53, May, 1990 49

the loop instruction overhead, I got
around 675,000 MAC/sec. When I put
the code in internal RAM, the speed im-
proved by 50,000 MAC/sec to over
725,000 MAC/sec. Now we're talking!

Note: You can download the code for the
matrix vector multiply and the code for per-
forming the transposed vector matrix mul-
tiply (needed during the error correction
phase of the neural network simulation) from
the Micro C BBS. The transposed code isn't
quite as efficient as the first but still does
over 700,000 MAC/sec.

Whew! I think we've eked every bit
of speed out of the transputer. You can
use this code with both online and epoch
training. Take your PC code and modify

sors) increase linearly in performance
with an increase in processors.

One such approach is a "processor
farm." In a processor farm one proces-
sor serves as the "farmer" and the re-
maining processors serve as "workers."
The worker processors execute identical
code. The farmer processor sends pack-
ets of data to each worker. When a
worker finishes with its work (its
packet), the farmer retrieves the results
and sends a new packet. Farming con-
tinues until the work is done.

This approach is probably the most
appropriate for using multiple transput-
ers to simulate neural networks. Each
worker runs its own neural network.

weights and the weight corrections
through the pipeline. Each pipe proces-
sor has a complete copy of the neural
network's nodes as well as the input and
output values for each exemplar vector
that it's responsible for.

The weights are stored and modified
in the administrator processor at one
end of the pipeline (typically another
transputer). The administrator processor
is connected to the host PC and to the
first and last processors in the pipeline.
It gets the topology and exemplar and
test vectors from the PC (typically in
files) and then uses this information to
configure the pipe processors.

When the pipe processors have their

Figure 5 — Processors

Arranged in a

Pipeline

Host PC

Administrator
Processor

#1 Pipe
Processor

#2 Pipe
Processor

#3 Pipe
Processor

#N Pipe
Processor

it to insure that the data for the weight
matrix is arranged correctly. Otherwise,
code written for the PC using Turbo C
(from Borland International) or Micro-
soft C compiles under the Logical Sys-
tems' C compiler with little or no
modification.

Using Transputers In Parallel

So far, I've used a single transputer to
simulate neural networks. If you want to
process more data and/or process data
as quickly as possible, then a parallel ap-
proach to transputing is the ticket.

The underlying architecture of the
transputer, with its multitasking and
synchronized communications capabili-
ties, almost eliminates the programming
hassles associated with controlling
several tasks concurrently.

Processor Farms

To implement a problem on several
transputers, you need to distribute the
workload as evenly as possible.

Some algorithms fit more naturally
into a parallel implementation than
others. So, you look for algorithms
which (when applied to parallel proces-

Pipelining

While a processor farm makes sense
once you know the neural network's
weights, it's not good for training. An al-
gorithm called pipelining is a better ap-
proach for training.

Pomerleau et. al. describes the algo-
rithm for use with the Carnegie Mellon
University Warp machine (a very expen-
sive systolic array processor). Chong
and Fallside 6 describe it for the trans-
puter.

This algorithm uses a group of pro-
cessors arranged in a pipeline (see Fig-
ure 5). In a pipeline, each processor
receives intermediate results and /or
data from its upstream neighbor,
processes the data, and then sends its re-
sults/data to its downstream neighbor.

For best results, try to communicate
(send /receive data) and compute
(process data) in parallel. Also, each pro-
cessor's work load should be relatively
equal. So no processor waits for another.

For neural network simulations on an
array of transputers, you typically send
a subset of the exemplar vectors to each
pipe processor. Then you pipe the

exemplar vectors and a copy of the
neural network nodes, the administrator
processor sends a copy of all the weights
down the pipeline. As a pipe processor
receives the weights, it immediately
sends them to the next processor
downstream. Then (in parallel) it uses
the weights to make a pass through the
neural network for each of its exemplar
vectors.

The pipe processor compares the cal-
culated output values with the exemplar
output values, then calculates an error
value which it sends back to the admin-
istrator processor. The administrator
node receives the error values from all
the pipe processors and determines if
the total error for all the exemplar vec-
tors meets the desired criteria.

If the error is low enough, the admin-
istrator processor transfers the weights
back to the PC and runs through any test
vectors that have been supplied.

If the error is too high, the adminis-
trator sends a message to each pipe pro-
cessor instructing it to calculate a change
of weights based on its exemplar vec-
tors. The administrator processor then
sends an empty array down the pipeline.

50 MICRO CORNUCOPIA, #53, May, 1990

When a pipe processor receives the
array, it adds the change in weights that
it's calculated to the array, then sends it
downstream to the next pipe processor.

When the administrator processor
gets the array back, it will have the sum
of all the changes for each weight based
on all the exemplar vectors. The array is
then multiplied by the learning rate con-
stant and added to the previous changed
weight vector (which has been multi-
plied by the desired momentum con-
stant) to produce a new weight vector.

The administrator creates a new set of
weights by adding the present weight
vector to the new weight vector. Then
the whole process begins again with the
administrator node sending the new
weights down the pipeline.

Programming Transputers

To implement the pipelined neural
network simulation on the transputers
requires only three programs, no matter
how many processors are involved. The
first program (ADMIN) is the adminis-
trator processor program.

This program interfaces to the host

PC and to the pipeline. It's responsible
for:

(1) Retrieving the topology of the
neural network;

(2) Transmitting the information
down the pipeline so that each pipe pro-
cessor can configure itself;

(3) Retrieving all the exemplar vec-
tors from the host;

(4) Transmitting evenly divided sub-
sets of the exemplar vectors to each pipe
processor;

(5) Transmitting the weights into the
pipeline;

(6) Getting error measurements from
each processor in the pipe;

(7) Determining whether the total
system error is below a user specified
value;

(8) If necessary, instructing the pipe
processors to calculate the weight
changes;

(9) Sending an empty weight change
array downstream;

(10) Collecting the filled weight
change array from the pipe processors;

(11) Using the host (i.e., user) sup-
plied momentum and learning rate con-

stants along with the weight change
array to calculate a new weight vector;

(12) Looping back to step 5;

(13) When training is complete,
saving the weights;

(14) And, if necessary, running test
vectors on the trained neural network
and collecting statistics.

The program (PIPE) is run on all the
pipe processors. It's responsible for:

(1) Receiving and retransmitting the
topology of the neural network;

(2) Setting aside enough space for the
weights and its own exemplar vectors;

(3) Reading and storing its own ex-
emplar vectors;

(4) Receiving and retransmitting any
exemplar vectors destined for processors
downstream;

(5) Receiving and retransmitting the
weights;

(6) Using the weights to calculate the
output for the neural network for each of
its exemplar input vectors;

(7) Comparing the neural network
calculated outputs with the exemplar
output vectors and calculating an error
value;

Save "Man-Years of Effort" with Turbo 5.5

Don't Start from Scratch
with Object-Oriented Pascal

Object Professional is a huge library
of over 200 object types and 2000
methods that
will multiply
your produc-
tivity. Window
object types let
you use over-
lapping and
resizeable
windows. The

windows include ■ scrolling data
entry screens ■ pick lists ■ menus

■ file selection ■ printed forms

■ help capability and more.

Build your programs using
proven data object types like stacks,
linked lists, virtual arrays, and more.
System-oriented routines provide
swappable TSRs in only 6K of RAM,
EMS management, and much more.

Satisfaction guaranteed or your money back within 30 days.
Add $5 per order for shipping in U.S. and Canada. Inquire
about other shipping charges. OPro requires Turbo 5.5.
BTF requires Turbo 4.0, 5.0, 5.5, or QuickPascal.

Object Professional includes
clear, comprehensive documentation,
on-line help, full source code,
technical support, and hot demo
programs. Pay NO royalties. You'll
get up to speed fast with OOP!

* 'The range of objects is
fantastic. Object Professional
could literally save you
man-years of effort. "

Jeff Duntemann

Object Professional 1.0,
only $150.

A Multi-User B-Tree Toolkit

Write powerful network
compatible databases faster and
easier using B-Tree Filer 5.0.
You'll have the fastest, safest, most
flexible databases - no rigid struc-
ture, no TSR hassles, no running out

of files. And they're compatible with
Novell, 3Com, MS-NET, and others.

You get ■ Fixed and variable
length records ■ Two billion
records per database ■ Up to 100
indexes per index file ■ Fail-safe
mode with journaling ■ Units for
sorting, browsing, reindexing, and
network control.

B-Tree Filer includes full source
code, documentation, technical
support, and you pay NO royalties.

** B-Tree Filer... a well rounded,

feature-rich approach to

B-Tree databases. "

Computer Language, 1/90

B-Tree Filer 5.0, only

$125. (single user)
With network support, $175.

* Call toll-free to order.
1-800-333-4160

8AM - 5PM PST Monday through Friday, USA & Canada.

For more information call (408) 438-8608. Fax: (408) 438t8610.

TurboPower Software PO Box 66747 Scotts Valley, CA 95067-0747

Reader Service Number 194

MICRO CORNUCOPIA, #53, May, 1990 51

(8) Transmitting its error value
downstream;

(9) Receiving and retransmitting any
error values produced by upstream pro-
cessors;

(10) Receiving and retransmitting any
instructions regarding a backward pass
through the network;

(11) If necessary, using the output er-
rors from all the exemplar vectors to cal-
culate the changes in the weights;

(12) Receiving the weight change
array from upstream, adding the locally
calculated changes, and transmitting the
array downstream;

(13) And looping back to 5.

The third program (TAIL) runs on the
last transputer in the pipeline in parallel
with a copy of the PIPE program. Unlike
the other processors that used the hard-
ware DMA channels (links) to transfer
data from one processor to the next, the
PIPE and TAIL programs on the last
processor communicate with each other
over a memory channel.

From a programmer's perspective,
memory channels operate identically to
hardware channels. This means that the
last PIPE program is identical to the
PIPE programs on the other processors.

The TAIL program primarily trans-
mits information coming out of the pipe-
line back to the administrator processor
and collects garbage. When the neural
network topology information reaches
the last pipe processor, there's nowhere
to send the data.

Since all the PIPE programs are iden-
tical, the last PIPE program will try to
send the data on anyway. TAIL receives
the topology data and discards it. The
weights transmitted down the pipeline
are likewise garbage collected. The error
values and the weight change array,
however, are sent back to the adminis-
trator.

Each of the processors has a harness
program. The harness programs specify
whether specific channels are hardware
or memory channels and which pro-
grams (ADMIN, PIPE, or TAIL) are to be
run together. The TAIL program can run
on the same processor as the ADMIN
program instead of with the last PIPE
program. Or, if space allows, a copy of
the PIPE program can run on the admin-
istrator processor. To maintain load
balance, it might be necessary to reduce
the number of exemplar vectors that this
first PIPE program processes.

Finally we must write a short con-
figuration file (used by the network

loader program) to indicate which pro-
gram is to run on which transputer. The
network loader is responsible for reset-
ting all the transputers and loading each
one with its specific program. From the
file, the network loader can tell which
transputers and which links a program
must be transmitted through before it ar-
rives at the transputer on which it's to
(ultimately) run.

Unfortunately, programming trans-
puters isn't always as simple as I've
made it sound. Fortunately, vendors are
improving their operating environ-
ments.

For example, some vendors support
message handling protocols. With some
of the protocols, a communications
process runs continuously in parallel
with the application program. Any time
data must be sent to another transputer,
the application program uses a vendor
supplied function or a memory channel
to pass the data to the communications
process. This process adds a header to
the data and sends the message out the
appropriate link.

Another transputer receiving the
message uses the header to determine
whether the message is intended for
another transputer or for itself. If neces-
sary, the process forwards the message
off to another processor. Otherwise the
process strips the header off the message
and sends the data through a memory
channel to the appropriate application
process. While this scheme slows down
applications, it can make communica-
tions simpler.

More important, these communica-
tions processes give you a means for de-
bugging remote transputers. Although
the application program may have
stopped on a remote processor because
needed data hasn't communicated to it,
the communications process can still
send and receive messages. One of those
messages can be an instruction to read a
particular memory location or register
and send the data back over a link.

Discussion

The topology of the neural network,
the number of test vectors, and trans-
puter memory resources will determine
how you manage the transmission of
weights. In most situations the adminis-
trator transputer transmits the weights,
storing them in the pipeline processors.

During the error correction phase of
the training cycle, each processor gener-
ates changes for the weights. These

changes are then sent back to the admin-
istrator where momentum and learning
rate adjustments get made. This ap-
proach requires few (if any) parallel
processes.

If, however, the network is very large
and there are memory limitations at the
nodes, you might have to break the
weights into packets (say all the inbound
weights for a hidden unit) and send
them sequentially down the pipeline.

The PIPE program would then calcu-
late all the partial sums for each exem-
plar vector. With all the weights for one
layer received, the activation function
would be calculated for each unit.

There are, of course, other ap-
proaches to dividing a neural network
simulation among transputers. For ex-
ample, reference 7 describes how each
transputer can process a subset of the
nodes. Other researchers have used
transputers with non-back-propagation
neural networks. 8,9,10

In Sum

The back-propagation simulation of
neural networks maps very nicely into a
parallel processing environment. While
other parallel processing platforms are
available, transputers provide a cheap
way to optimize neural network simula-
tions. Plus, we can use these transputers
for other applications as well.

References

1 Pelczarski, M., "System Review:
Microsoft Softcard," BYTE, vol. 6, no. 11,
November, 1981, pp 152-162.

2 Inmos Limited, The Transputer Refer-
ence Manual, Prentice Hall, London,
1988.

3 Atkin, Phil, "Performance Maxi-
misation," Technical Note 17, Inmos-
SGS Thomson, Bristol, UK, March 1987.

4 "Some Issues in Scientific Language
Application Porting and Farming Using
Transputers," The Transputer Develop-
ment and iq Systems Databook, Inmos Ltd.,
1989.

5 Pomerleau, D. A., Gusciora, G. L.,
Touretzky, D. S., Kung, H. T., "Neural
Network Simulation at Warp Speed:
How We Got 17 Million Connections per
Second," Proceedings of IEEE Inter-
national Conference on Neural Net-
works, San Diego, USA, July 1988, pp
11-143-150.

6 Chang, M. W. H., and Fallside, R,
"Implementation of Neural Networks
for Speech Recognition on a Transputer
Array," Technical Report CUED/F-IN-

52 MICRO CORNUCOPIA, #53, May, 1990

FENG/TR8, Cambridge University, De-
partment of Engineering, Cambridge,
UK, March, 1988.

7 Beynon, T., "A Parallel Implemen-
tation of the Back-Propagation Algo-
rithm on a Network of Transputers,"
Research Initiative in Pattern Recogni-
tion, Royal Signals and Radar Estab-
lishment, Malvern, UK, Poster Paper,
1988 IEEE International Conference on
Neural Networks.

8 Di Zitti, E., Caviglia, D. D., Bisio, G.
M., and Parodi, G., "Neural Networks
on a Transputer Array," Proceedings of
the 1989 International Conference on
Acoustics, Speech, and Signal Pro-
cessing, pp 2513-2516.

9 Abbruzzese, F., "A Transputer Im-
plementation of a McCulloch & Pitts
Network," Parallel Processing and Ap-
plications, E. Chiricozzi and A. D'Amico
eds., North-Holland, 1988, pp 135-140.

10 Board, J. A., Jr., and Lu, J. S. J.,
"Performance of Parallel Neural Net-
work Simulations," Proceedings of the
Second Conference of the North Ameri-
can Transputer Users Group, J. A. Board,
Jr., ed., Durham, North Carolina, USA,
North American Transputer Users
Group, 1989, pp 185-200.

11 Lee, D. G., Jr., "Preliminary Re-
sults of Applying Neural Networks to
Ship Image Recognition," Proceedings
of the 1989 International Joint Confer-
ence on Neural Networks, June 18-22,
1989, Washington DC, IEEE, San Diego,
11-576.

12 Rumelhart, D. E., Hinton, G. E.,
and Williams, R. J., "Learning Internal
Representations by Error Propagation,"
in Parallel Distributed Processing Volume
1: Foundations, Rumelhart and McClel-
land ed., Cambridge, Massachusetts,
USA: MIT Press, 1986, Ch. 8, pp 318-362.

13 Leung, H. C, and Zue, V. W.,
"Applications of Error Back-Propagation
to Phonetic Classification," Proceedings
of the Neural Information Processing
Systems - Natural and Synthetic Confer-
ence, November 28 - December 1, 1988,
D. S. Touretzky, ed., Morgan-Kaufman,
1989.

Vendors

Companies advertising Transputer
TRAMs and /or language compilers for
use with the Transputer:

Inmos-SGS Thomson
1000 Aztec West
Almondsbury Bristol BS12 4SQ
UK

Levco

6181 Cornerstone Ct. East, Ste. 101

San Diego, CA 92121

Multis Corporation
99 Willie Street
Lowell, MA 01854

Micro Way, Inc.
P.O. Box 79
Kingston, MA 02364

Computer Systems Architects
950 North University Avenue
Provo, UT 84604

Logical Systems
P.O. Box 1702
Corvallis, OR 97339

3L Limited

Peel House

Ladywell Livingston EH54 6AG

Scotland

Sension

Denton Drive

Northwich Cheshire CW9 7LU

Transtech Devices Ltd.

Unit 17, Wye Industrial Estate

London Road

High Wycombe

Buckinghamshire HP11 1LH

Companies advertising neural net-
work software using coprocessor cards
and IBM PC compatible computers:

SAIC

10260 Campus Point Dr.

San Diego, CA 92121

HNC

5501 Oberlin Drive

San Diego, CA 92121

Company advertising neural network
software for Transputers and Apple
Macintosh II computers:

Neurix

One Kendall Square, Suite 2200

Cambridge, MA 02139

♦ ♦ ♦

You're in charge

Don't be limited by someone
else's idea of the perfect menu. Su-
perMenu's unique graphical interface
adds a level of creativity, individual-
ity, and ease-of-use not found any-
where else. Why shoehorn your
software into a stuffy, preconceived
mold when SuperMenu gives you
complete control over the structure,
format, and appearance of your
menu? Most menu systems give you
a modest choice of screen colors and
then claim to be customizable. Super-
Menu gives you color plus a palette
of drawing tools, multiple fonts, fill
patterns, line styles, pop-up notes,
and more. You can even edit the
background pattern. In fact, Super-
Menu has many features you'd only
expect to see in an expensive drawing
program, much less a menu generator.

Cut Copu Pasta Front Back Group Una roup Align Seal*

I" [fflfff] Wolfe ffiano &bop fffTfffl -|

■ I I I I I II I MASTER MENU I I I I I I I I ■

3 H

•jr

Susan's Clients

{SALES PEOPLE"] | PUBLISHING | | CLIENTS | | MUSIC |

Rar-lnipR I,,™"" 1 , w )

[ BACKUP - MONDAY |

| BACKUP - TUESDAY |

| BACKUP - WEDNESDAY |

j BACKUP - THURSDAY |

| BACKUP - FRIDAY |

SuperMenu's easy-to-use Because SuperMenu has a

mouse driven interface makes wide range of powerful tools,

running your favorite software a menus can be as complex or as

snap. Other computer tasks, simple as you desire. Like we

such as daily backups, can be said, you're in charge,

easily automated and accessed — c\/7

with a simple mouse click. ^tB|p(l^]Ml(MjQBCi

Graphical mouse-driven
interface

Instant access to your soft-
ware and DOS options
Works wM popular programs
Easy-to-use drawing
tools, fonts, fill patterns,
line styles, and colors

Multiple menus, each
with up to 99 pages
Hypertext-like links
between pages
Pop-up notes
Passwords prevent un-
authorized modifications
to your menus

Date and time display
NOT memory resident
Enter DOS commands or
invoke a DOS Shell
Supports CGA, EGA,
VGA, and Hercules
graphics

4995

Add $5.00 S & H

Please specify 5V4" or

3VS" when ordering

Requires: IBM PC, XT, 286, 386, PS/2 or compatible. DOS 2.0 or greater. 256K RAM.
Microsoft, Logitechorcompatiblemouse.CGA.EGA, VGA, Herculesorcompatibledisplay.

Reader Service Number 216

1293 NW Wall Street, Suite 71
Bend, Oregon 97701

(503) 389-5489

MICRO CORNUCOPIA, #53, May, 1990 53

Save The Floppies!

R x For Slipped Disks

This is probably the easiest to use disk
recovery program I've ever seen. You just
insert the flaky disk and fire up the program.
When it finishes, every sector on the disk
will be readable, probably.

Data error reading drive B
Abort, Retry, Ignore?
Grrrrr...
Back in the old days, when CP/M
had trouble reading a floppy, I could
break out the dependable and user-
hostile program DU (Disk Utility) for a
little rescue work. DU didn't care about
bad sectors; it would read with reckless
abandon. Often, the "bad sector" had
just a single bad byte. An easy repair job
would restore the file's readability.

I thought it would be useful to have
this ability in an MS-DOS system. Forget
for the moment that MS-DOS already
lets you repair disks with its RECOVER
program. At the beginning of this pro-
ject, I hadn't made the acquaintance of
RECOVER.

A Little Background

Each sector on a disk consists of sync
bytes, ID address marks, ID fields (cyl-
inder, head, sector, bytes per sector), a
CRC value for the ID fields, a gap, more
sync bytes, data address marks, data
(whew, at last), a CRC for the data, and
yet another gap. That's at least 105 bytes
of overhead for each 512 bytes of data.
Add another 146 (or so) bytes of header
info for each track and you'll get an idea
of what the Floppy Disk Controller
(FDC) has to wade through.

At the lowest level, the FDC, acting
under orders from the CPU, selects a
drive, fires up the drive motor, seeks to
the track, performs the requested func-
tion, and reports the results to the CPU.

During a read, the FDC looks at the

assuming that a
CRC error means a
data error, and
barring physical
damage to the disk,
simply rewriting the
questionable data
to the same sector
should (and has, on
numerous
occasions) restore
the file.

track until it either finds ID fields match-
ing those requested, or sees the disk's
index hole pass twice without finding
the matching fields (sector not found). If
it finds the sector, the FDC sends the 512
bytes of data to the host — in MS-DOS
systems, to the Disk Transfer Address
(DTA).

As the FDC scans sector ID fields, it
generates a CRC value for the IDs and
compares this with the CRC recorded on
the disk. If the two don't match, the con-
troller returns a CRC error.

A second type of CRC error comes
when the CRC calculated during the
data read doesn't match the data CRC
recorded on disk. This is the error we're
after. It means that one or more bytes of

data have changed (or that the CRC has
been corrupted). Why'd they change?
Beats me: a stray magnetic field, a glob
of peanut butter, cosmic rays. . . .

It would be nice to isolate the data
CRC error from the ID field CRC error.
But after a wade through the XT BIOS
listing, I can't see that it differentiates be-
tween the two. So we can either make
the rash assumption that any CRC error
will come from the data, or we can dive
into the FDC to get a closer look at the
error.

Leave your trunks at home (leave
your rash, too). It doesn't make sense to
try to fix the sector ID since it gets writ-
ten only during formatting; reformatting
a track would wipe out the information
we want to recover. We probably
couldn't read the sector anyway, since
the corrupted ID fields wouldn't match
those requested by the CPU.

Assuming that a CRC error means a
data error, and barring physical damage
to the disk, simply rewriting the
questionable data to the same sector
should (and has, on numerous occa-
sions) restore the file.

Levels Of Disk Access

Most programs use DOS interrupt
21 h to read a floppy. DOS deals with
files, so application programmers like it.
No fuss, no muss. But we'd like to get
into some real trouble here. We can get
at the whole disk by dropping one level
closer to the hardware and making use
of the BIOS.

Nestled in ROM, the BIOS doesn't
know or care about directories, files, or
any of the other high level nonsense that
DOS requires. With the BIOS you can
access any sector on any disk, whether
if s allocated to a file or not (including
the directory and FATs). And, like DOS
INT 21 h, you can choose to ignore error
conditions.

54 MICRO CORNUCOPIA, #53, May, 1990

Larry Fogg

187 NW Jefferson PI.
Bend, OR 97701

Figure 1— FLOPFIX.C

/* FLOPFIX.C - Reads all sectors of a floppy. Any sectors with bad
CRCs are rewritten. */

#include <stdio.h>
#include <dos.h>

int crc_failures;

int sectors_recovered;

int non_crc_f ailures ;

void reset FDC ()
{
union REGS in, out;

/* reset the floppy disk controller */

in.h.ah = 0;
int86 (0x13, Sin, Sout)
} /* reset_FDC () */

/* service = reset */
/* INT 13h */

unsigned char read_sector

{

union REGS in, out;

(unsigned char track, unsigned char side,
unsigned char sector)

in.h.ah = 0x2f;

intdos (Sin, Sout); /*

/* get DTA */
do the INT, ES:BX holds address on return */

in.h.ah = 2;
in.h.al = 1;
in.h.ch = track;
in.h.cl = sector;
in.h.dh = side;
in.h.dl = 1;

int 8 6 (0x13, Sin, Sout) ,
return (out .h . ah) ;
} /* read_sector () */

/* read diskette */
/* # sectors to read */

/* drive B: */

/* do the read */

/* return read status */

unsigned char write_sector (unsigned char track, unsigned char side,

unsigned char sector)

union REGS in, out;

in.h.ah = 0x2f;

intdos (Sin, Sout) ; /*

/* get DTA */
do the INT ,ES:BX holds address on return */

in.h.ah = 3;
in.h.al = 1;
in.h.ch = track;
in.h.cl = sector;
in.h.dh = side;
in.h.dl = 1;

int86 (0x13, Sin, Sout),
return (out . h . ah) ;
} /* write_sector () */

/* write diskette */
/* # sectors to write */

/* drive B: */

/•do the write */

/* return write status */

void fix_disk ()

continued on page 56

The key here is that, even though an
error condition exists, the data has been
read. It's alive and well at the DTA, and
all you have to do is rewrite it to disk.

Code Notes

You won't find anything very start-
ling in FLOPFIX.C (see Figure 1). fix-
diskO makes three attempts to read each
sector. It takes a while for the drive
motor to spin up, so a well behaved pro-
gram should do these retries to ensure
that any errors are real. (Reads from a
half-fast drive don't have much chance
of success.) A call to reset_FDC() follows
any read failure — again, a required pro-
gramming practice.

Once we're sure an error has oc-
curred, fix_disk() checks the status re-
turned by read_sector(). If the status
indicates a CRC error, write_sector() at-
tempts to rewrite the sector. Global vari-
ables for the number of bad sectors and
recovered sectors get updated and we
move onto the next sector.

You'll notice that each sector access
(in write_sector() and read_sector())
starts with a call to determine the DTA.
This could have been done only once
during program initialization, since the
DTA doesn't change during execution
(unless the program changes it expli-
citly). But the call to DOS function 2Fh
provides a convenient way to load
ES:BX with the DTA.

But Why Bother?

Sure, sure. Once again, I've rein-
vented the wheel. But in some situations
it's a better wheel, a bit more round. RE-
COVER has not been as successful as
FLOPFIX in restoring damaged sectors
to readability (at least, in my humble
opinion). On one occasion a RE-
COVERed file was not TYPEable. Be-
sides, diving into the guts of the
machine keeps me out of trouble. It's al-

MICRO CORNUCOPIA, #53, May, 1990 55

ways instructional to try to understand
the inner workings of a system.

If I had more gumption, I'd disas-
semble RECOVER to see what makes it
tick. I could rip off the good stuff and
add it to FLOPFIX. But wisps of cloud
over the nearby Cascades promise a fine
day of early spring back country skiing.
Some other time for the disassembly.

Improvements

As usual, I've kept things very
simple. Any number of obvious addi-
tions would make FLOPFIX better.

Adapt FLOPFIX to hard drives. No
real problem, but hard drives are a little
scary.

Use DOS to make the program file
oriented rather than sector oriented. This
would let you recover a specific file.

Write the recovered sector to a differ-
ent disk (or, at least, to a different sec-
tor). This would allow for recovery of
physically damaged sectors.

I see nothing wrong with keeping it
simple and easy to use. Dave's always so
complimentary about my code. I believe
his words were something to the effect

'C and 'C++'

DOCUMENTATION TOOLS

Save TIME and MONEY
Automate ACCURATE documentation

• C-CALL™ ($59) Creates a graphic-tree
of the caller/called structure, generates
a files-vs-functions table of contents, and
a cross-reference of function usage.

C-CMT "($59) Generates and inserts
function "Coment-Blocks" showing
caller/called functions and showing
global/define/local identifiers used.

C-LIST H($39) Lists programs with
optional line numbers and page titles.
Optional graphic "Action-Diagrams" of
the logic/control structure. Optional
reformat into standardized formats.

> C-REF ^$49) Creates cross-reference
of global/define/local identifiers.
Creates Class-Hierarchy-Diagram of
C++ Classes (incl. multiple inheritence).

SPECIAL OFFER!! ($149) All 4 programs
plus FREE integrated C-DOC™ program.
« 30-DAY Money-back guarantee !

SOFTWARE BLACKSMITHS INC

6064 St Ives Way,

Mississauga, ONT, Canada. L5N-4M1

(4l6)-858-4466

Reader Service Number 219
56 MICRO CORNUCOPIA, #53, May, 1990

Continued from page 55

unsigned char track, side, sector, status;
char tries;

reset_FDC () ;

for (track=0; track<40; track++)
for (side=0; side<2; side++)

for (sectoral; sector<10; sector++)
{

tries = 1;

status = read_sector (track, side, sector) ;

if (status !=0) /* error on first read */

{

do /* try 2 more times */

{

reset_FDC (); /* always reset after an error */

status = read_sector (track, side, sector) ;
tries++;
} while ((tries<4) £& (status !=0) ) ;
if (status !=0) /* bonafide error */

/* do 40 tracks */

/* 2 sides */

/* 9 sectors */

/* first attempt */

{

reset_FDC () ;

if (status=0xl0)

/* reset after failed read */
/* is it a CRC error? */

crc_failures++;

if (!write_sector (track, side, sector)) /* write OK */

{

sectors_recovered++ ;

printf ("CRC error tr=%d side=%d sector=%d OK\n",
track, side, sector) ;
I
else

/* couldn't rewrite sector */

{

resetJFDC (); /* reset after failed write */

printf ("CRC error tr=%d side=%d sector=%d BAD\n",
track, side, sector) ;
)
}

else
{

non_crc_failures++;

printf ("NonCRC error=%xh tr=%d side=%d sector=%d\n" ,
status, track, side, sector) ;
)

/* not a CRC error */

}
} /* fix disk () */

main ()

{

crc_failures = non_crc_failures = sectors_recovered = 0;

puts ("\nFloppy repair program");

puts ("Put disk in drive B: and hit a key...\n");
. getch (); /* wait for key press */

fixjdisk () ;

printf ("\ncrc failures detected = %d\n", crc_failures) ;

printf ("sectors recovered = %d\n", sectors_recovered) ;

printf ("non-crc failures detected = %d\n", non_crc_failures) ;
}

that, "Any idiot could use this pro-
gram." True. I leave it to the intrepid
reader to add features and move FLOP-
FIX beyond the "idiot phase."

Fini

So much for my contribution to the
final issue of Micro C. It has been a very
great pleasure to meet you all in these
pages over the last four years. Perhaps
we'll bump into each other again in

another magazine, or pawing through
some bin hidden in the bowels of an out-
of-the-way surplus parts house. 'Til
then....

semper ubi sub ubi

♦ ♦ ♦

WORLD

By Laine Stump

% Redhouse Press
Merkez PK 142
34432 Sirkeci
Istanbul, Turkey

Batch File Prophylactics
And C++ 2.0

Laine tries out Zortech's new C++ compiler and
debugger at 2 a.m. Here's his red-eye special on this
new package.

Just a note here at the beginning for those of
you who think you're underpaid: An article
I read last week in the Turkish daily news-
paper Sabah (Morning) reported that, according
to a new pricing schedule issued by the Turkish
government, the wages of an unskilled worker
are now lower than the wages of a donkey. A
donkey working as a beast of burden is now
paid 480,000 liras ($202) per month, while a
farm or factory worker is paid 225,000 ($95).

Editor's note: Wonder how much supervisor
donkeys get.

This is especially comic in a country where
the blank in the common phrase "You son of a
&*%#@!" is replaced with "Donkey." (The way
things are going, that may soon become a com-
pliment). On the other hand, it is frightening
when you consider that food for an average
family of four is twice what the worker receives.
And that's not counting rent, electricity, water,
transportation....

News like this really makes me appreciate
the (relative) security of the U.S. or Western
Europe.

Not that I think you aren't underpaid. Just
thought it would make you feel better if you
knew it could be worse.

C++ 2.0

My newest toy these days (it just arrived last
Friday) is Zortech's C++ 2.0 Developer's Edi-
tion. I've been up until 3 a.m. the last couple
nights experimenting and reading. I've used
Zortech 1.07 for the last year and wanted to find
out what they had and had not fixed. So far
most everything looks good.

The C++ Developer's Edition includes not
just a C++ (native code) compiler, but also a
similar C compiler, a Source level C++ (!) de-
bugger, and a library of C++ classes to use in
your own programs.

Also included is a TSR library that allows
you to make your own programs into popup
TSRs with (apparently) not too much work. I
will refrain from saying anything good or bad
about this until I've tried it. Two days just isn't
enough time to sort through all these goodies.

As with just about every compiler on the

MICRO CORNUCOPIA, #53, May, 1990 57

market, they've included an editor
(ZED). Since I've become hopelessly
hooked on Brief, though, I haven't done
much except accidentally call it up once
or twice from within the debugger. Be-
cause of this, I can't say anything to rec-
ommend or condemn it.

And, of course, there is the plethora
of utility programs. A Linker, MAKE,
TOUCH, OBJ2ASM, ARCHIVE, UP-
DATE. Even a MAKE file dependency
generator. Everybody has to have a try
at making a better wheel.

C++ Compiler

The Zortech C++ 2.0 release adds all
the new features of the C++ 2.0 defacto
standard (defined as "what is contained
in AT&T's cfront 2.0") except for a few
additions to the stream I/O library
(which are "currently proprietary to
AT&T"). The two most important addi-
tions to the language in version 2.0 are
multiple inheritance and type safe link-
age.

Multiple inheritance means that a
derived class can inherit characteristics
of more than a single base class. You
might find this useful if, for example,
you wanted to make a derived class Car
that uses the facilities of both classes
Body and DriveTrain. (Why you would
want a class called Car is beyond me.
Maybe you really need the class called
JunkHeap which uses Smoke and Rust
base classes.) You can see from this ex-
cellent example why multiple inheri-
tance is useful.

Type safe linkage means that the type
information of classes, variable, and
functions is carried not just in the *.H
files, but in the object file as well. This
way, you avoid the following scenario:

(1) compile srca.cpp which uses
lib.hpp for type information;

(2) modify lib.hpp and lib.cpp to
change a function prototype;

(3) recompile lib.cpp;

(4) link srca.obj with lib.obj.

The problem is that if all type check-
ing is done at compile time (as happens
in C and C++ 1.0), it is possible to link
two modules which have different ideas
of what type certain pieces of data and
functions are. This could lead to dis-
astrous, unexplainable results.

In C++ 2.0, type checking is done not
only at compile time, but also at link
time. This is implemented in Zortech
C++ (and AT&T cfront 2.0) by adding
type information to the name of each
variable (the compiler). That way if the

type information is different, you will re-
ceive an "unresolved reference" error
from the linker.

C Compiler

In terms of its use as a C compiler,
Zortech 2.0 has quite a few changes as
well. Of course all these changes apply,
by definition, to the C++ compiler, too.

MS C (MS Windows) Compatibility

Possibly the most important change is
all the work done to make Zortech com-
patible with MS C (and Turbo C). You
can now use Zortech C++ to develop MS
Windows, OS/2, and OS/2 Presentation
Manager programs. (You will also need
to have the Windows Developer's Kit
and/or the OS/2 Softset. The OS/2 ver-
sion of the compiler is a separate up-
grade, not included with the Developer's
Package.)

Achieving this compatibility re-
quired, among other things, adding sup-
port for the "pascal," "near," and "far"
keywords, and updating the code gener-
ator to generate the special procedure
prologs and epilogs required by Win-
dows.

There are a few incompatibilities,
though. In MS C, normally SS==DS (al-
though this must be changeable with an
option to compile OS/2 and Windows
DLL libraries, among other things). This
is not the case in large data memory
models of Zortech C, which put SS in a
different segment to allow more room
for global and static data.

This becomes a problem only when a
near pointer is declared in a program
where data is normally far. The near
pointer is always assumed to point to
DS; in MS C it can be used to point to an
automatic or parameter variable (since
SS==DS). It cannot in Zortech C, accord-
ing to the C++ Compiler Reference chapter
called "Converting Microsoft C Pro-
grams to Zortech C/C++."

Another problem is that Zortech does
not support the HUGE data type; you
can't have any single data object larger
than 64K. I have always avoided HUGE
arrays anyway (because of their inherent
inefficiency), so that isn't a problem for
me. You may not like making linked lists
of arrays, or arrays of pointers to arrays,
however.

Although Zortech has done a lot of
work to ensure compatibility between
their library and the Microsoft library,
there's still a problem with open(). It
"takes different parameters and behaves

differently than the Microsoft C ver-
sion," according to Zortech.

I should point out, though, that just
because Zortech' s library is compatible
with MS C's, it doesn't mean you can
link Zortech OBJs with the MS library, or
use third party libraries compiled for MS
C! This is almost always not the case.
Compatibility between Zortech and MS
C is at the source (function prototype)
level.

With some work, it is possible to com-
pile a library that can be linked either
with MS C or Zortech OBJs, but not
many companies do it that way. Most
have a separate version of their library,
which is simply the same source code re-
compiled with the other compiler.

Finally, the "interrupt" function type
available in MS and Turbo C is not avail-
able in Zortech. Instead they supply an
Interrupt Package with their library
which handles the installation of inter-
rupt service routines.

Having used both methods, I prefer
Zortech' s; it doesn't force you to play
around with saving the old vectors or
worrying about whether to chain to the
old interrupt. It will even set up a local
stack for the interrupt if you like. Much
handier.

There are a few other incompatibili-
ties, all detailed in the manual. The ones
I have listed are the most important,
however. Most of the rest are just due to
the fact that MS C 5.1 is not fully ANSI
compatible. So programmers are forced
into writing non-ANSI compatible code,
which Zortech C doesn't like.

The big advantage of switching from
MS C to Zortech is that now you can
write your Windows and PM programs
in C++ instead of mere C.

Compiler Performance

There's no reason for me to write
about this. You can find all the numbers
in the advertisements. I will say that the
speed of everything is more than accept-
able (I never was much for numbers)
and the object modules Zortech creates
are maybe 2% smaller than previous ver-
sions.

Whatever the numbers, they are de-
finitely within Developer class. This is
especially true for running the compiler,
since it generates OBJs directly, rather
than requiring the extra step of translat-
ing C++ to C, as most other C++ pack-
ages.

Of course, the optimizer takes some
time. It can be lived with, though. Just

58 MICRO CORNUCOPIA, #53, May, 1990

use it once after all debugging is done.

With Zortech 1.07, I once mistakenly
reset my machine, thinking the op-
timizer had crashed. The program was
one I had semiautomatically converted
from Pascal, and it was filled with array
subscripts and a 500 line case statement.
Being used to non-optimizing compilers,
after 20 seconds of no disk activity (I'm
impatient), I assumed the machine had
crashed. Later I figured out my mistake.
I also rewrote the program.

C++ Debugger (ZDB)

Several issues ago, I did a mini-re-
view of the original Zortech C Debugger.
It was a promising product, but lacked
many features I needed to make it my
only debugger. The biggest problem was
that I couldn't use it to debug assembly
language programs.

In the ensuing months, rather than
keep CodeView on my disk and switch
constantly from one debugger to the
other, I simply began to write more pro-
grams in C and C++, and fewer in as-
sembly. When I did write an assembly
language program that I needed to
debug, I used good old SYMDEB (the de-
bugger included with MASM until ver-
sion 5.0).

Finally, with the release of ZDB, I
have a single debugger for everything.
As well as testing ZDB with several of
my C programs, I also tested it with two
programs written in 8086 assembly lan-
guage and assembled with MASM 5.1. It
even worked on a program for which I
had no source.

Debugging In General

The original version of ZDB (called
ZTCDB) could only handle C programs.
While many of the improvements to the
new version involve support for C++
and assembly language debugging, there
have also been changes to many other
parts of the program.

ZDB as a C Debugger is very reminis-
cent of ZTCDB, but with some nice addi-
tions. For starters, although ZDB still
needs its debugging (symbol, line num-
ber, variable type) information in a
different format than the standard
CodeView information produced by
most linkers, it now automatically does
the translation rather than requiring you
to run an extra program (ZTCMAP). It
also now saves this information at the
end of the EXE file rather than in a sepa-
rate file.

Another nice change: when debug-

ging C programs in assembly mode
(where source file lines are displayed in-
terspersed with the assembly language
instructions generated by the compiler),
you can single step by assembly lan-
guage statements. In the old version, you
could only step from one source state-
ment to the next, skipping over many as-
sembly instructions in the process.

Also, you no longer need to worry
about confusion between multiple static
variables with the same name. These are
displayed in the data window as "varia-
ble#file," where variable is the name of
the static variable and file is the source
file which contains it.

Another very convenient feature (if
you can stand learning a new editor), is a
command which puts you in the editor
(ZED), editing the file currently in the
source window with the cursor on the
currently selected line. That makes fixing
bugs much easier. Unfortunately, there is
no facility for connecting this to any edi-
tor other than ZED.

There were cosmetic changes as well.
For example, previously the menus were
Lotus style (on two lines at the top); now
they are pull down menus, making it
much easier to see the available options.

But, as with most large, complex pro-
grams, there are still Cs & Bs (Com-
plaints and Bitches).

First, ZDB assembly mode still dis-
plays locals and parameters (e.g., word
ptr [BP-6]). On one hand, it is nice to see
just how the variable is being accessed. (I
can remember complaining about
CodeView doing just the opposite be-
cause it concealed what was really hap-
pening.)

On the other hand, it confuses the
issue. All the information is there (in the
CV debug information) to display this
properly. Why not at least make it an op-
tion? For example, it could be displayed
in raw form when in the display mode
where instruction bytes are listed to the
left of the disassembly, and in symbolic
mode otherwise.

Also related to assembly mode dis-
play is the habit of ZDB to believe that
the last few assembly statements of the
last function in a source file are actually
in "No File." There is a kind of No Pro-
gram's Land between each source file.

Another problem, which only occurs
if you use SEC (Someone Else's Com-
piler) or SEA (Someone Else's Assem-
bler) is that ZDB has no idea what to do
with 186 opcodes. This is unfortunate,
since MASM, among other programs,

generates some 186 opcodes automat-
ically (if you tell it to do so). These op-
codes are displayed as "????" in the
source window.

Yet another bug: after accidentally
going into ZED (the editor) a few times
when I didn't want to, I erased it from
the \ZORTECH\BIN directory. I figured
that ZDB would now give me an "Editor
Not Available" error. Much to my sur-
prise, it instead went into an endless
loop of scrolling the screen up each time
I pressed a key, with no apparent possi-
bility of escape. I finally ctrl-alt-del'ed
out of it.

It would be especially useful in ZDB's
dual monitor mode (then I could turn all
the lights down and have an office that
looked just like Gordon Letwin's office
on the cover of Inside OS/2). Of course
ZDB does support 43 and 50 line by 80
column modes, but only on EGA and
VGA cards. If they would just give me
one day with their video library
source....

Anyway, none of the above problems
hamper normal operation. Overall, ZDB
is much improved over the original. Just
as I did when I first experimented with

Bar Coding Source Code

BAR CODE IN C

Barcodes are now in
widespread use. Don't let
your application be left out!

Do you or your customers need a faster and
more accurate method of data entry? Does
your application need to print barcodes?

The SymBCG collection of C routines
supports UPC (A & E), EAN (8 & 13), Code
39, 2 of 5 (Industrial & Interleaved), and
Codabar symbologies. Routines are
included for printing on HP LaserJet and
Epson dot-matrix printers.

Full source code for all symbologies are
included. No royalties.

Keyboard wedge barcode readers are also
now available.

Only

$99.00

Symbologic

953 E. Colorado Bl. #342

Pasadena, CA 91 106

(818)449-7114

BONUS: Order SymBCG and a "MiniReader" barcode reader
for $350 and save $50!

California residents please add 6.75% sales tax.

Foreign orders, P.O.'s, C.O.D.'s please call.

Reader Service Number 202
MICRO CORNUCOPIA, #53, May, 1990 59

CodeView, I spent at least an hour last
night watching the call stack and local
variables bip up and down as it stepped
through recursive calls to FactoriaK).
Maybe it will even be useful, as well as
fun.

New Debugging Modes

As mentioned above, ZDB can now
debug not only C programs, but also As-
sembly and C++ programs, and those
with no source at all. The only thing it
can't do is debug a packed EXE file. (The
EXEPACK utility included with MASM
compresses these files. Nobody else can
directly debug them either.)

Sadly, it also can't debug a smelly cat.

Debugging Without Source

When ZDB is asked to debug a file for
which it can find no source, or no debug
information, it starts up in raw assembly
mode. That means it works like a
stroked-and-bored DEBUG, in a screen
oriented fashion (with a memory display
window, disassembly window, register
window, conditional and unconditional
breakpoints, tracepoints, etc.)

Debugging Assembly With Source

To test the ability of ZDB to debug as-
sembly language, I used the FACT.ASM
and FACT2.ASM programs which I had
given as examples when I talked about
MASM 5.1 and CodeView in Issue #45.
These programs were written using Pas-
cal calling conventions (arguments are
pushed left to right, rather than right to
left as in C). They still loaded into ZDB
and executed with no problem. Almost.

Even things that the ZDB manual said
wouldn't work went without a hitch. For
example, I had declared a variable to be
FAR PTR WORD (yes, MASM can do
that). The manual said this wouldn't be
displayed properly in the Data window
(it would display the value of the pointer
as an integer, rather than the value at the
address pointed to by the pointer).

Instead, I was cheerfully greeted by a
display of the pointer in segment:offset
form and, after pressing the Ins key, was
shown the value at the location it
pointed to. The contents of local varia-
bles (automatic, stack) displayed cor-
rectly, too.

The only rough spot was that ZDB
did not correctly recognize any function
with no parameters or local variables.
For example, when I called up FACT,
which has no locals in main(), ZDB
started in assembly display mode rather

than source display mode. I switched to
source mode and hit the Single Step key,
only to see the program branch off to no-
where. (ZDB did not, however, branch
off to nowhere. It politely allowed me to
reload the program).

The fact that ZDB expects all func-
tions to set up the BP register to point to
their local stack frame (local variable
space) causes this problem. For some
reason, it gets confused about which
source line goes where if this isn't done.
Unfortunately, MASM does not set up
the BP register if a function has no locals
or parameters.

I tried doing the same thing again,
this time leaving the display in assembly
language mode, and it stepped through
the program just fine. It also worked per-
fectly in source mode with FACT2,
which has a local variable in main().

When ZDB loads a program, it looks
for the symbol _main (or main). If it
finds that symbol, and if the normal

PUSH BP
MOV BP,SP

instructions are at that address, it starts
debugging in source mode with the in-
struction pointer (IP) sitting on the in-
struction just after the above stack frame
setup code. If it doesn't find one of these
conditions (or if you specify the /a
switch), it starts in assembly language
mode with IP sitting at the very first in-
struction to be executed.

Starting at main() is useful for C and
C++ programs. That way you skip over
all the library startup code (which is sup-
posedly already debugged anyway). I
usually don't call the starting point in an
assembly language program main,
though. That is, I didn't used to.

Now, if I want the debugger to start
in source mode, I must name my main
routine "main." Not only that, but I
must assemble with the /mx switch to
turn on case sensitivity, since ZDB thinks
that MAIN and main are different (and
rightly so).

Another problem with FACT and
FACT2 was that, as mentioned above,
ZDB doesn't recognize 186 opcodes. I
had originally set the programs to use
186 opcodes (with the .286c directive).
This caused MASM to use the ENTER in-
struction instead of the function prolog I
showed above. ZDB displayed this as
"????." I removed the .286c and reas-
sembled, eliminating the problem.

These aren't big problems, though

(with the possible exception of the non-
recognition of 186 opcodes). If all I must
do is name the start address of my pro-
gram "main" and use /mx to debug as-
sembly language with ZDB, I can live
with it. Happily.

Debugging C++

This is the real exclusive of ZDB. Al-
though other companies may have
source level C++ debuggers in develop-
ment, or maybe even secretly on the
market, Zortech has the first source level
C++ debugger for MS-DOS that I've
seen.

You may think that any multi-lan-
guage debugger should be able to debug
C++ just as well as C, Pascal, or any
other language. That's kind of true. For
example, I can compile a C++ program
with Zortech and load it into CodeView.
It will step through the program with no
problem.

But just try looking at a function
name! Say you have a class called Car
with member function Crash(int mph,
long *damage). You would expect the
label for Crash to be _Crash. Not so! In-
stead, the label is _Crash 3CarNipl.

The "3" tells how many letters are in the
class name, "N" means it is a near func-
tion, and "ipl" are the argument types
(integer, and pointer to long). Obviously
this could get tedious.

Also, while many debuggers have
facilities for dealing with struct data, no
others can properly handle instances of
classes. As an example, if you expanded
an instance of Car, you might see the
value of Model, but not of Engine (if En-
gine is really a member of the base class
DriveTrain).

ZDB has special C++ debugging sup-
port to handle these problems. Most im-
portant in this support is name
unmangling, which simplifies function
name display, and a new window: the
Class window.

The Class window allows convenient
probing through an interrelated set of
class definitions. Consider that each
class definition may be a derived class
based on several other class definitions,
which may themselves be based on
several other classes.... The class win-
dow helps to sift through these connec-
tions and, if desired, put up the source of
a member function in the debugger's
source window.

Of course, the Class window just
shows definitions of classes and member
functions. You still must use the Data

60 MICRO CORNUCOPIA, #53, May, 1990

Figure 1—KBD.C

/* kbon.c - a program to enable the keyboard and optionally
execute a command, then disable the keyboard.

written by Laine Stump Jan ??, 1990
No Eights Reserved.

# include <stdio.h>
#include <ctype . h>
#include <process.h>
finclude <dos . h>
# include <int.h>

extern int _okbigbuf - 0; /* these reduce resident program size */
extern unsigned _stack = 1024;

void kbon (void)

{ /* enable input from the keyboard */

outp (0x64, OxAE) ;
#ifdef DEBUG

puts ("kbd ON") ;
#endi£

} /* kbon() */

/* enable keyboard command */

void kboff (void)

{ /* disable input from the keyboard */
unsigned ct;

/* clear out keybuf */
/* wait for key release */

/* disable kbd port */
/* clear kbuf again to be sure */

while (bioskey (1) ) bioskey (0) ;
for (ct =0; ct < 65535; ct++)

outp (0x64, OxAD);

while (bioskey (1) ) bioskey (0) ;
#ifdef DEBUG

puts ("kbd OFF");
#endif

} /* kboff {) */

'/* - */

int breaktrap (struct INTJDATA *pd)

{ /* return means "call original handler" */

return (1); /* 1 return means "don't call original" */

) /* breaktrap () */

-*/

/* — -

int get kbf lag (char *arg)

{ /* process the string at arg to determine if we want the
keyboard turned disabled (/d) , or enabled (/e)
Returns for /d, 1 for /e, and -1 for error */
if (*arg++ =='/')

if (toupper(*arg) — 'D')

return (0) ;
else if (toupper(*arg) — 'E')
return (1) ;
return (-1) ;
) /* getkbflag() */

/* */

void usage (void)
{

puts ("usage: kbd /e|/d [command] \n" ) ;

puts(" where /e enables keyboard input, /d disables keyboard" );
puts(" input, and the optional 'command' is a dos command to");
puts(" execute while keyboard is enabled/ disabled . If command");
puts(" is specified, the keyboard is disabled/enabled after the") ;
puts(" command is executed. \n") ;
exit(l);
} /* usage () */

/* */

int main (int argc, char *argv[])
{

int retval, kbf lag;

if ( (argc < 2) || (kbf lag = getkbflag(argv[l]) ) = -1)

usage {) ;
int_intercept (0x23, breaktrap, 256); /* make sure we don't get */
if (kbf lag) kbon(); else kboff (); /* ctl-Breaked before we */

and Automatic (local variables) win-
dows to examine the contents of in-
stances of a class. The functionality of
these windows has been upgraded to
allow the same traversing operation to
see the values of member variables of a
class instance.

C++ Tools

The C++ Tools include classes for
queues, singly and doubly linked lists,
stacks, bit vectors, virtual arrays, hashed
search tables, windows, and all kinds of
other things. Included is a book which
explains each class, gives examples of its
use, and gives the complete source
listing for the class (the source is on disk
as well). If nothing else, the C++ Tools
are good examples for those just starting
with C++.

I had wanted to play with these
classes some and get a feel for how use-
ful they were. But, as with the TSR pack-
age, there just wasn't enough time. Rain
check.

Documentation

The old version of Zortech had a
single manual for the compiler and li-
brary and another manual for the de-
bugger. This was more or less adequate,
but I still found myself reaching for my
old Turbo C 1.0 manual now and then
(especially in the old days, when I was
weaning myself from Pascal — it has a
great little section showing equivalent
operations in C and Pascal side by side).

The manual set for 2.0 has a separate
manual for the compiler and utilities, the
libraries, the debugger, and the C++
Tools. I've done quite a bit of thumbing
through them in the last two days and
have found most of the information I
needed. There is even a thorough ex-
planation of symbol construction in C++,
and many other implementation details
that almost nobody would need. But
when they did....

There are sections both on AT&T C++
compatibility and MS C compatibility. It
also has a section titled "Incompatibili-
ties with ANSI C," which deals not with
incompatibilities between Zortech and
ANSI, but with incompatibilities be-
tween any C++ and ANSI C. This is very
useful reading for a new convert from C.

Price

Unfortunately, the $29.95 JRTs and
$49.95 Turbos are a thing of the past.
Well, not really. I've been seeing JRT ads
again recently but, not able to believe,

MICRO CORNUCOPIA, #53, May, 1990 61

thought that I had entered a strange time
warp like one of those episodes from
"Twilight Zone." Seriously, I thought
the magazine had misplaced some of the
copy from their April issue.

The problem is that the amount of
development (and the amount of adver-
tising dollars) that has gone into this
kind of software has skyrocketed since
the days of JRT and Turbo 1.0.

Even Borland realized long ago that
$49.95 per copy wasn't even keeping the
lights on, much less paying for the
$10,000 BYTE ads and financing all those
hot tub parties and hotel suite bashes so
essential in modern (marketing) warfare.
The price of Turbo Pascal has risen with
each new release (it seems like the up-
date fee for the new edition is about the
same as the retail price of the previous).
I stopped at 3.0.

The retail price of the Zortech
Developer's Package 2.0 is $450. The
C++ compiler by itself is $199. Prices
from discount houses will probably run
considerably less.

On the other hand, the functionality
of JRT Pascal, or Turbo Pascal isn't even
close to the functionality of Zortech C++.
Although considered breakthroughs in
their time, they are now mere toys.

Also, the retail price of Microsoft C
5.1 is about the same. And it's just a C
compiler, no C++ (it does include OS/2
support, however, which is a $150 up-
grade from Zortech).

Still, it isn't even fair to compare Zor-
tech C++ with MS C. Or any C, for that
matter. Because it isn't C. It's C++. And
it has a debugger that is light years
ahead of the last version I've seen of
CodeView.

Even if the price was a bit higher
(which it isn't), it would be worth it in
all the development time saved. After
all, how many days does it take you to
make $450? Don't you think it's about
time you worked a few nights so you
could afford to have some fun?

Zortech Inc.

1165 Massachusetts Ave.
Arlington, MA 02174
Voice: (617) 646-6703
Fax: (617) 643-7969

Batch File Prophylactics

Before closing this thing up and toss-
ing the smelly cat out for the night (you
thought I was just kidding about smelly
cats), I want to pass along a little "cute-
C" utility I wrote. I compiled it with

if (argc > 2) /* get a
{ /* execute program */

retval = spawnvp(0, argv[2] , 6argv[2]); /*
if (kbflag) kboff{); else kbon();

shance to kboff()*/
execute command */

else

retval - 0;
int_jrestore(0x23) ,*
return (retval) ;
} /* main */

/*
/* send back

restore DOS ctl-Break
exit value of command

♦ ♦ ♦

Zortech, but it will work with MS or
Turbo with just a few changes in places I
have pointed out in the source. See Fig-
ure 1.

KBD

KBD is a program that can enable
and disable the keyboard on a PC com-
patible. This can be useful when writing
a BAT file which you don't want the
stupid user to screw up, or running a
program which you don't want the....
You get the idea.

The syntax of KBD is:

kbd /e|/d [command]

/e enables the keyboard, /d disables
the keyboard, and command is an op-
tional parameter, which is a DOS com-
mand to execute while the keyboard is
enabled /disabled. If command is given,
the keyboard is turned back off/ on
when the command finishes. For ex-
ample:

kbd /d

turns the keyboard off. If you type this at
the DOS prompt, the only way you can
get the machine going again is the reset
button or the Big Red Switch! Usually
you would do it at the beginning of a
sensitive batch file. Another example:

kbd /d chkdsk

kbd /d brot ...

(I thought Larry might appreciate
that second one.) These commands turn
the keyboard off while executing chkdsk
(or brot), then turn the keyboard back
on. As I said, KBD is mostly useful in
BAT files (especially AUTOEXEC).

As an example, say that you have a
BAT file that gets a user's name and

password with the command LOGIN
before continuing. For example's sake,
LOGIN returns if the login was
successful, or 1 otherwise. You want to
put LOGIN in a BAT loop that looks at
errorlevel. The danger is that the user
can simply type ctl-C and break out of
the batch file. This is where KBD comes
in. For example, you might use this BAT
file:

echo off

break off

kbd /d

:loop

kbd /e login

if errorlevel 1 goto loop

kbd /e

break on

Notice that, although "break off"
minimizes the number of times DOS
checks for ctl-C, it doesn't eliminate it.
That is what makes KBD essential. KBD
also traps the DOS ctl-C interrupt (23h)
while it is executing the command,
making sure nobody can ctl-C out of that
program.

Of course, you have the problem that,
while the first two lines of the BAT file
are executing, the keyboard is still
enabled. A good way to solve this is to
make a device driver that turns off the
keyboard. This way the keyboard is dis-
abled before the user has any chance to
screw things up. I'll leave that one as an
exercise for the reader.

How?

KBD turns the keyboard on and off
by outputting a command to the key-
board controller port (64h). OADh turns
off the keyboard, and OAEh turns it on.
This physically disables anything from
the keyboard. Even ctl-alt-Del falls on
deaf ears.

Interrupt Package

Another interesting part of KBD is

62 MICRO CORNUCOPIA, #53, May, 1990

the use of the int_intercept function from
the Zortech interrupt package. The inter-
rupt package handles all the details of
saving the original interrupt vector, put-
ting the new one in place, setting up a
stack for the service routine, chaining to
the old service routine if requested, and
restoring the old vector when finished.

Note that breaktrapO (my replace-
ment for the ctl-C interrupt) returns a 1.
This means "don't chain to original in-
terrupt." If it returned a it would mean
"chain to the original interrupt now,
then return." Since our entire reason for
trapping the interrupt was to keep the
original from being called, we don't
want to return 0.

DOS Exit Values

In the example above, we used error-
level to check the exit value of the
LOGIN program. But the program being
executed (by the BAT file) is KBD, not
LOGIN! How can I do that? With the
clever trick of using one of the spawnO
functions (spawnvpO in this case), which
returns the program's exit code. Then I
return that value from mainQ in KBD

(the return value of mainO is sent to DOS
as the exit value of the program).

spawnvpO searches the DOS path for
the command and can use the arguments
in argv[] without modification, but it
cannot call a BAT file. If I wanted the
ability to execute BAT files, I could use
systemO instead, but then I wouldn't get
exit values. (systemO uses com-
mand.com, which eats the exit value.)

_okbigbuf and _stack

The declarations of the variables _ok-
bigbuf and _stack at the top of KBD.C
are included to reduce the resident size
of the program. This leaves more
memory for the program called with
spawnvpO.

In Zortech C, when _okbigbuf is 1
(the default), the maximum 64K of data
space (for small data models) is allocated
on program startup. If _okbigbuf is 0,
data space is only allocated from DOS as
needed. _stack controls the size of the
program's stack. By setting these two
variables, I was able to trim the resident
size of the program from 72K to 10K. A
big improvement for two lines!

XT SCHEMATIC

Micro Cornucopia

Phone Orders:
(503) 382-5060, M-F, 9 AM-5 PM PST

Mall Orders:
P.O. Box 223, Bend, Oregon, 97709

IBM PC-XT Schematic

.$15.00

As always, you can do anything you
like with KBD. Even swear at it if you
want. I simply don't care.

Pseudoepidermatologistical
Meanderings

Sorry, but it's nearly 2 a.m., and you
can't expect everything I say at 2 a.m. to
be coherent. The column is finished now.
But it's cold out, and the smelly cat looks
so pitiful. (If she just wasn't about to ex-
plode with babies.) I guess she can stay
inside tonight. But just this once. (Any-
body want a Turkish kitten?)

♦ ♦ ♦

PROMPT DELIVERY!!!

SSAME DAY SHIPPING (USUALLY)
QUANTITY ONE PRICES SHOWN for FEB. 25, 1990

OUTSIDE OKLAHOMA: NO SALES TAX

1MB

SIMM

1Mbit

41256

4464

41264'

DYNAMIC RAM

MEM DeskPro 386/20
AST Prem386/33Mhz

1Mx9

256Kx9

1Mx1

256Kx1

256KX1

64Kx4

80 ns

100 ns

80 ns

60 ns

80 ns

100 ns

120 ns

100 ns

27C1000
27512
27256
27128

EPROM

128Kx8 200 ns

64Kx8 200 ns

32Kx8 1 50 ns

16Kx8 250 ns

STATIC RAM

62256P-10 32Kx8 100 ns

6264P-12 8Kx8 120 ns

l6116ap-12 2Kx8 120 ns

$295.00

225.00

92.00

35.00

8.75

4.15

2.50

2.20

1.95

2.50 1

7.50

$18.00 1
7.80
6.50]
3.75

$10.00
4.50
4.25J

OPEN 6 DAYS, 7.30 am-10 pm. SHIP VIA FED-EX ON SAT.

SAT DELIVERY MasterCard/VISA or UPS CASH COD

included on MICROPROCESSORS UNLIMITED, INC.

FED-EX ORDERS 24,000 S. Peoria Ave., /Q4 Q\ OC7 4QCH

RECEIVED BY: BEGGS, OK. 74421 \W I Of C\J I "430 I

flu S-2 $&25/4b No minimum Order. Please note: prices subject to change!

Fr: M S1&50/1 lb Shipping, insurance extra, up to $1 for packing materials.

DEVELOP YOUR OWN

BOOTABLE
OPERATING SYSTEM

WITH BOOTER TOOLKIT,,

Real-time scheduler

DOS-compatible file system

Window system

Memory manager

Bootstrap loader

100+ page user manual

ONLY $99 WITH SOURCE CODE

ORDER HOTLINE

(206)391-4285

For immediate shipment, send your
check or money order for $99 to:

GENERAL
SOFTWARE-

General Software
P.O. Box 2571 , Redmond, WA 98073

Telephone orders shipped UPS BLUE COO only. To order by credit card, call Programmer's Connection or Austin Codewords,
authorized General Software dealers. Second-day air shipping provided tree to destination only within the United States. Foreign
orders add $20 00 lor airmail delivery. Washington Residents add 8.1% sales tax. Copyright (c) 1990 General Software. All rights
reserved. Booter Toolkit and General Software are trademarks of General Software.

Reader Service Number 37

Reader Service Number 212

MICRO CORNUCOPIA, #53, May, 1990 63

64 MICRO CORNUCOPIA, #53, May, 1990

YOUR OWN

You're On Your Own

By David Thompson
Laura Shaw &
Nancy Ellen Locke

Micro C Staff

Well, here it is: probably for the first time in
the history of computers, an article in a com-
puter magazine on advertising in other com-
puter magazines. So many advertisers have
asked for suggestions of where to advertise that
we've put together this information.

Who's Offering The Best Deal?

The standard way of judging what kind of
deal you're getting on advertising is to divide
the one-time B&W full-page price by the circula-
tion (in thousands). That's the cost /thousand.
That way you can compare rates from smaller
magazines with rates for the big guys.

Once you figure the cost/ thousand readers,
you can factor in the less precise variables. What
percentage of the audience would be interested
in your product? How much training would
they require? How much will you have to ex-
plain in the ad? (You can say ICE to designers;
you have to explain what it means to managers.)
Would they pay what you're asking?

Be aware that corporations look for image
and support, hackers look for performance and
price.

But when you're all done, the real question
is: Will advertising your product in Computer
Language or Programmer's Journal or PC Tech-
niques be profitable? If you make money in all
three, then advertise in all three. If your product
doesn't make money anywhere, maybe you
need a new product.

As for what to expect, check out "On Your
Own" in Issue #51. Don and Kim Jindra's ex-
perience advertising their $25 network should
give you a good idea what happens.

My advice for prospective advertisers is:

(1) Figure out how much space you need to
describe your product. If you're selling some-
thing that's easy to describe, you don't need
much space. If it's a totally new idea, or a sub-
stantial improvement on an old idea and re-
quires a lot of explanation, plan to buy a lot of
space.

(2) Figure out how much image you need to
buy. (This is very different from #1.) Let me give

you an example: if you're selling a $1,500 CAD
package, be prepared to buy full-page four-color
ads and find the best ad agency to produce
them. Your ads must create a very solid image
for your product. If you're selling a $25 net-
work, a simple micro ad may draw more orders
than any other size.

(3) Make sure your ad gets seen. (This
sounds a bit like #2.) If you only need a tiny ad
(a cheap product that's easy to describe), you
may find that a micro-sized ad grouped with
other micro ads may be better read than a quar-
ter page hidden among the articles. If you need
at least a half page to tell your story and you're
afraid of getting lost, then go for a full page. Full
pages are very visible. Four-color full pages are
most visible.

Also, if you're buying a half page or more,
you might as well act like an experienced adver-
tiser. Ask for a right-hand page, outside corner,
and ask them to place it with a particular
column or article. (If it's a very large magazine,
ask them to place it in the front third). They'll be
glad to charge you extra to guarantee you a par-
ticular position, but ad people can often finagle
you a visible spot, no extra charge, if you simply
mention you'd like it.

Plus, you may as well tell them you're acting
as your own agency. That way you'll get the
standard 15% agency discount, off the top.

(4) Does the ad have to make the sale? Not
always. You can send out a lot of information in
a 1-ounce letter. Just make sure your ad is as
good at weeding out weak prospects as it is at
attracting strong ones.

(5) Something other than ad space? There's
direct mail. Figure on spending anywhere from
$.75 to $1.50 per piece. (That counts label, en-
velope, return envelope, literature, postage, de-
sign....) Also figure on a 2% to 5% response, so
you'll spend from $15 to $75 on each order.

Generally, magazine ads are a faster,
cheaper, easier way to reach a lot of people.
Mail works best as a follow up to enquiries
generated by your ads and for selling additional
products to recent purchasers.

MICRO CORNUCOPIA, #53, May, 1990 65

Now That You're Going To Advertise

(1) Order media kits. The kits contain
the latest ad rates (rates can change from
one month to the next), ad deadlines, ad
sizes, how they should be produced. . . .

(2) Watch the deadlines. If you ask for
space, you'll probably have to pay for it
even if you don't get your materials to
them on time. Stay on top of your ad. It
may be obvious to you and me that ads
need to change as products change. But
you'd be surprised how many small
companies run outdated ads for six
months simply because they keep
missing deadlines.

(3) Make sure your ad fits. A lot of
people send in incorrectly sized ads.

(4) Make sure it looks good. You want
solid black type, clean lines, white paper,
no smudges, no grey, no dirt.

(5) Keep a second copy of your ad (a
high-quality, printable copy). Remem-
ber, Murphy works for the U.S. Post Of-
fice and his brother runs Federal
Express.

(6) Check everything for accuracy
(particularly the phone number). I know
more than one old lady who's been inun-
dated with orders for hard drive utilities.

(7) If your product runs under $100
and your ad's doing the selling, you may
not want a reader service number. I
know several companies who have been
crippled by the costs of replying to
reader service numbers. (They assumed
most of the recipients would respond;
turned out few, sometimes none,
ordered.)

(8) Talk to the magazine's editor. Ask
him if he thinks you'd do well advertis-
ing there. Also, ask him if there's some-
one you should send a copy to. (Don't,

absolutely don't, mention that your deci-
sion to advertise depends on his decision
to review the product. You won't feel
good about it and he won't either.)

Once The Ad Has Run

(1) Ask everyone who calls where
they heard about your product.

(2) Make sure you can respond imme-
diately to orders. We started Micro C out
of our house and we lived on a dead-end
road. We'd meet the mailman as he
stopped at our box and many times we
had reply cards addressed and ready to
hand back to him by the time he'd gotten
back to our house. We had 10-minute
turnaround and subscribers loved it.

(3) Be prepared to lose money and
patience on some orders. Occasionally
you'll run into a real bother. He'll want a
custom product, he'll want you to install
it, he'll call you day and night instead of
reading the manual, he'll erase his disks;
and after you finally return his money,
he'll write a nasty letter to your state's
attorney general suggesting that the
death penalty might be too lenient. (The
guy was from New York City, of course.)

Note: Circulation figures are, at best,
optimistic. When you're on the phone
with a new magazine, ask for a break-
down of their circulation. (Ask the ques-
tions in the following order.)

(1) Is your circulation audited by ABC
or BPI? If so, great; have them send or
fax you the latest publisher's statement
and skip the following questions.

(2) How many paid subscribers in the
most recent issue? (Not next issue, or the
end of next year.)

(3) How many copies of the most re-
cent issue were sent to the newsstands?

(4) What percentage of the newsstand

DEVELOPMENT LIBRARIES FOR PC PROGRAMMERS

illlll M IIII H Iiiiililllll . ilil H iillllillllll . illll j iilll H llll j

product perfanguqge a&tanggages

♦ BTree-core and disk versions $42,95 $139,00

♦ lists, stacks and queues $34.95 $109.00

♦ Tree$— binary and heaps $42>96 S 139.00

ail packages $96.00 $295.00

order direct;

REGAN SALES
VISA PO BOX 2204

AM£XPR£$$ (206)620-2603

copies normally sell? (40% to 50% is nor-
mal, 60% is high, 70% is almost unheard
of)

(5) What was their total print run for
the most recent issue?

Listen for pregnant pauses, hemming
and hawing, fudging, explaining, etc. If
the totals of #2 and #3 add up to #5, then
they're probably fudging. Either way,
take printed circulation figures with a
giant grain of salt.

Also, don't be afraid to call a ma-
gazine's advertisers. They'll tell you a
lot. Some magazines are comfortable to
work with, others aren't. Some will
make deals, others won't. Sometimes
companies find the best response in
brand new magazines, other times they
do best in the old timers. In some publi-
cations the response is immediate, in
others it takes three ads. Sometimes ad
placement is very important, other times
it isn't.

You'll be amazed how much informa-
tion you can glean in half a dozen calls.

The Carrot

Of course, there's no feeling like
opening your post office box and seeing
it full. Really full. Tends to make all the
hassles, all the preparation, worthwhile.
Reminds me of a little known saying
that's been handed down from hacker to
hacker over the centuries.

"May your bugs be gentle, may your
CRT be warm upon your face, and may
your mailbox be ever filled with checks.
Good checks."

Writers Market

Speaking of checks in the mailbox,
there's more than one way to squeeze
cash out of a computer. If writing soft-
ware isn't your bag, then how about arti-
cles? The most effective writers:

(1) Have a specialty: processor de-
sign, graphics algorithms, database diag-
nostics, entrepreneurial experience. . . .

(2) Have a conversational style: An
open, easy style rather than a stiff, aca-
demic, jargonizing (agonizing) presenta-
tion.

(3) Write for the magazines they keep
in their bathrooms.

(4) Don't take themselves too seri-
ously. I'm talking about both their deal-
ings with editors, and the way they
write.

I wish you well, my friends.

♦ ♦ ♦

Reader Service Number 210
66 MICRO CORNUCOPIA, #53, May, 1990

MARKETING ALTERNATIVES

Magazine

Publisher

Address

Contact

Phone

Circ.

Ad Rates

The C Gazette

Andrew Binstock

1341 Ocean Ave.
Santa Monica, CA
90401

Andrew Binstock

213-473-7414

6,000

Full

1/2
1/4

page $450
page 230
page 130

Midnight
Engineering

Bill Gates

111 E. Drake
Ft. Collins, CO
80525

Bill Gates

303-491-9092

15,000
projected

500
250

175

PC Techniques

Keith Weiskamp

202 E. Greenway
Phoenix, AZ

80532

Jeff Duntemann

602-493-3070

20,000
projected

1,270
775
420

Tech Specialist

Robert Ward

2601 Iowa St.
Lawrence, KS
66047

Donna Ward

913-841-1631

10,000

1,150
575
290

The C Users
Journal

Robert Ward

2601 Iowa St.
Lawrence, KS
66047

Donna Ward

913-841-1631

27,000

1,150
575
290

Circuit Cellar

Daniel Rodrigues

12 Depot St.
Peterborough, NH
03458

Rose Mansella

203-875-2199

20,000

995
600
350

Programmer's
Journal

Liz Oakley

PO Box 31060
Eugene, OR
97403

Jay Moore

503-747-0800

30,000

1,895

1,195

645

Computer
Language

Regina Ridley

500 Howard St.
San Fransicso, CA
94105

Regina Ridley

415-397-1881

70,000

3,270
1,940
1,145

The C Gazette is a code-intensive C and C++ quarterly for MS-DOS
users. It specializes in in-depth articles aimed at the experienced C pro-
grammer. Code supports the major compilers and is accompanied by
clear, cogent prose. The C Gazette is committed to the exploration of
serious programming topics. Bruce Eckel of Micro Cornucopia has been
the Gazette's C++ editor for the last year.

Midnight Engineering, the journal of personal product develop-
ment, is the magazine for hardware and software developers who want
answers about developing and marketing their own products. Cover-
age includes just-in-time development, product pricing /marketing, and
tips /techniques for working "on your own." Published bi-monthly, a
one year ME subscription costs $19.95 for the introductory period.

PC Techniques is designed for a market of proven buyers of pro-
gramming languages, tools, application software, hardware, and more.
Readers are interested in the practical applications of programming re-
lated software products. Every issue will present hands-on software re-
views, programming tips, and practical advice from leading software
developers and authors.

Tech Specialist is written for advanced PC developers. This power-
ful group of professionals works in both software and hardware
development, using the stand-alone PC as a tool. Tech Specialist gives
PC developers a "real world" tool - integrating information about hard-
ware manipulation, software design, and the internals of the PC. Each
issue provides an in-depth analysis of specific development problems.

The C Users Journal provides an interactive forum for C program-
mers. The Journal publishes practical information for C programmers,

including advanced and intermediate programming hints and tech-
niques, tutorials, and information about software tools. Each issue fea-
tures major "how-to" articles written by experienced C programmers, C
product reviews, book reviews, correspondence from members, and in-
formation on new products.

Circuit Cellar INK, The Computer Applications Journal, is the
premier source of practical technical information for designers and
builders of computer hardware and software applications. Circuit Cel-
lar INK enhances the electronics design skills of its readers by offering
creative solutions and unique applications through complete projects,
practical tutorials, and useful design construction techniques.

Programmer's Journal publishes articles written by the industry's
top professional talent - working programmers who are willing to share
their valuable technical expertise for the benefit of fellow colleagues.
Every other month, professional developers worldwide go to PJ to find
language articles with useful source code on Ada, ASM, BASIC, C,
C++, COBAL, FORTRAN, FORTH, Modula-2, and Pascal - plus the
latest information on OS/2, the 80386/486, the EGA and VGA, new
product releases, industry news, and consulting tips.

Computer Language is the productivity magazine for professional
software developers. Its mission is to enhance the programmer's ability
to generate working applications on schedule. Each issue of Computer
Langague includes columns devoted to application design and analy-
sis, productivity tools, user interface design, object-oriented program-
ming, and "how to" articles written by the finest professional
programmers in the field.

MICRO CORNUCOPIA, #53, May, 1990 67

LGttGrS continued from page 6

figured it might not be the place for an
urban keyboard potato like me. I may
have been wrong, but about the rafting:
would I actually have to go outdoors
for that, or could you digitize the ex-
perience and send it to me on a floppy?
(By the way, does SOG stand for Sages,
Oracles, & Gurus or Silly Old Geezers?)
Anyway, thanks for putting out the
best computer mag around. And don't
give a second thought to that volcano
when you're trying to get to sleep. The
chances of it suddenly exploding into a
violent torrent of boiling lava and poi-
sonous gases and burying the entire
town under glowing cinders are really
small. Less than 1 in 10, probably.

M. N. Macleod
3043 S. Laredo Circle
Aurora, CO 80013

Editor's note: SOG stands for Anything
You Like — no, that's not it.... I suppose you
know that Bend almost closed when you
left. Fortunately, the other resident stayed.
As for the dead volcano, the humane society
picked it up last weekend. (The smell was
getting awful.)

Correction

As if the discussion isn't confusing
enough already, we managed to slip an
error into James Martin's letter on the
radar equation controversy (see Letters,
Issue #52).

At the bottom of the right hand
column on page 78, we refer to the total
power at the detector as proportional to
the inverse cube of R. Wrong. The cor-
rect term (as stated in James's original
letter) is the inverse fourth power of R.

How can this 25% power loss have
happened? Perhaps we're seeing proof
of the First Law of Preservation of Post-
al Power.

The PPP theory states that in all
Postal interactions, power is conserved.
In other words, the Postal Service's pro-
posed 25% increase in rates will be
balanced by a corresponding decrease in
all mail contents. Good news for all bill
payers, but be sure to tell your em-
ployer to stop sending your checks
through the mail.

We hope this will serve to clear up
the confusion and save James' sullied
reputation.

Micro C Staff

68 MICRO CORNUCOPIA, #53, May, 1990

More UNIX

Bob Morein's "UNIX Packages For
The PC" in Issue #50 was exactly the
type of information I've been looking
for. I turned up several other articles as
well as two useful books. (See my list at
the end of this letter.)

One of the things I'm discovering is
that UNIX is quite particular about the
hardware. For example, at least three
flavors of UNIX (and XENIX) do not
support RLL hard disk controllers, nor
do they support many video display
boards.

I haven't decided which UNIX I'll
purchase. That decision will have to
wait until I find out more about which
flavors support which hardware. Mean-
while, I hope you will have enough re-
sponses to Bob's article to encourage
more articles in Micro Cornucopia.

Fiedler, D., "Future Imperfect,"
BYTE, May 1989, p. 113.

Fiedler, D., "Calm Approach to
UNIX," BYTE, Aug. 1989, p. 113.

Combau, G., "The UNIX Shell,"
BYTE, Sept. 1989, p. 315.

Unger, J., "One Man's Experience,"
BYTE, May 1989.

Morein, B., "UNIX Packages for the
PC," Micro C, Nov.-Dec 1989, p. 40.

Fiedler, D., "UNIX on Personal Com-
puters: Why and How," BYTE, Sept.
1989.

Christian, K., The UNIX Operating
System, 2 n edition, John Wiley and
Sons, 1988.

Fiedler, D. & Hunter, H., L7MX Sys-
tem Administration, Hyden Books, 1986.

Larry Hoffman
10152 Oso Ave.
Chatsworth, CA 91311

Editor's note: Wonderful list, but you
missed the original tome on the subject: The
Hardware Requirements Of Eunuchs,
Xinu, MLVX BC.

Op-amp Expose

Dear Mr. (smug) Editor — Regarding
your comments to my letter in Issue
#52: I do not appreciate your conde-
scending and insulting response to what
was intended as a constructive correc-
tion. Turnabout being fair play, let the
games begin!

Obviously my letter isn't the only
thing you've ignored; OpAmps-101
must have occurred coincident with

your nap time. For your sake (as well as
others who have yet to tame the beast),
I shall endeavor to put into words the
secret of life, op-amps, and everything.

The op-amp is a device with two
(count them) inputs: one inverting (-)
and one non-inverting (+). It also has
one output (isn't that wonderful?).
Now, whatever happens on the non-in-
verting (+) input causes the output to
move in the same direction. Amazingly,
whatever happens on the inverting (-)
input causes the output to move in the
opposite direction. I guess that's why
they call it the inverting input.

The real secret (I know you've been
waiting for this) is: the op-amp will
swing its output, hoping against all
hope, waxing brave in this cold, cruel
universe, that somehow, somewhere,
someway there is a feedback path that
will bring its inputs into balance by
virtue of its efforts. This is usually done
by providing feedback to the inverting
(-) input. Yes friends, indeed, that little
booger won't rest until its inputs are at
equilibrium. The same. Identical. Twins.
You get the picture.

Knowing what we now know, let's
walk through the circuit as published
with a test case. Suppose we let
Vi=1.010 volts and V2=1.000 volts. By
the equation stated (Vout=Vi-V2) we
should expect V ou t to equal 0.010 volts.
The (-) input will be at V2/2 (0.500
volts). The (+) input has... well, it de-
pends where the output was to start
with. (Vi and V2 in the original article
should be swapped for the equation to
apply. Change the designations in either
the schematic or the equation, but not
both.)

Let's say the output was at 0.000
volts just a microsecond or two before
we applied the input. Since 0.000 volts
is as good as ground, the (+) input is at
Vi/2 (0.505) volts. The inputs aren't
balanced so the output will simply rise
until they are. (Ooh, we've got me
now!) There, see! The (-) input is at
0.500 volts, and the (+) input is uhm, er,
even bigger than it was before! (Just a
minute, let me grab my calculator.)
Nope, the inputs still aren't balanced.

"Hmmph" says Mr. op-amp, "I'll fix
that. I'll just raise my output even more,
and hope like crazy that somehow my
(-) input will rise to equilibrium with
my (+) input." Things are even worse;
the (+) input rises higher still.

"Now my inputs are really getting
out of balance. I'll show them!" Mr. op-
amp zings his output as high as he can
go. "Surely my inputs will be balanced
now."

Sadly, Mr. op-amp has pegged his
output, his (+) input higher than ever,
and his (-) input still with only 0.500
volts on it. Mr. op-amp does the only
thing he can do. He keeps on holding
his output high, for surely somewhere,
somehow, someway his (-) input will
rise, and he waits. He will be waiting a
long time.

Yes, op-amps are a bit more compli-
cated than that, but you won't stray far
from what's really happening if you
keep "the secret" in mind. Maybe if
anyone (else) cared, mine would not
have been the only letter pointing out
the mistake. After all, this is a magazine
about micros; who cares if you lie about
anything else?

What you presented in Issue #49 was
little more than a comparator (and a
poor one at that — too much hysteresis).
If you still don't believe me, build it.
Perhaps in your haughtiness, you, not I,
were thrown by the inverting (-) input
not being on top. In closing, may I sug-
gest you either stand corrected or sit in
shame.

Dave Stojan
10310 Lybert Rd.
Houston, TX 77041

Editor's cursory response: At first I was
kind of plus-minus about your letter, but I
like the way you amplified the subject.

Oh, Go Take A Flying @#%*

You very nearly blasted me off your
list of subscribers — back when you re-
marked that your Stinson had taken 30
years to become as obsolete as my Big-
Board II had become in three. I never
owned a Stinson, but I used to park my
7AC in a hangar next to one, so your
remark hit me with a zonk of 10 on the
Richter scale. However, many months
went by before I needed to renew Micro
C, and time is a notorious healer. I'm
still here.

I won't attempt to refute your re-
mark. But I'll soften the zonk ("Obsoles-
cence, where is thy sting?") by recalling
the following:

1. Mr. Alan Turing proved a theorem
which says that what one computer can

do, any computer can do. Of course, any
theorem is based on assumptions; and
unfortunately no theorem can make my
BBII do color graphics. Still, for compos-
ing this letter, my BBII is just as good as
the latest 32-bit screamer. A "Turing
theorem" for aircraft, on the other hand,
would have little practical value. There
simply aren't many tasks at which a B2
bomber and a Stinson are equally effec-
tive.

2. Unless your Stinson spent the 30
years in a museum, I'll wager the fabric
and most of the engine aren't original.
Someone has shelled out hard cash just
to keep that Stinson from becoming
worse than obsolete. My BBII has con-
sumed negligible cash.

3. If 3 BB years equal 30 Stinson
years, then 10 BB years is a century.
That spells "antique" where I come
from. So if I hang on a bit longer, I can
contest the high ground now held by
the Imsai.

Of course you'd have to dezonk your
Stinson analogies. To atone for past
excesses, you might even start a semi-
regular old folk's department. "The
Slow Lane," or "80s in the 90s," or "Is
There Life After Obsolescence," or "The
Trailing Edge." Something like that.

I guess I'm not your only reader who
wants to know where to load up on 8"
disks before antique dealers drive the
price up. Maybe there are others who
believe that good software won't appear
until the hotshots have moved on.

R. W. Hartung

408 Orchard

East Lansing, MI 48823

Editor's Hand-Typed Response: It
sounds like you're really up in the air over
my analogy, but I won't let that bring me
down. Would you accept a takeoff on flying
in place of an apology?

♦ ♦ ♦

See the difference
at a glance • . .

f textfile.old textfile.new N

hello

I world II

good

b£ZS

■boy

with

DELTA

THE COMPARISON TOOL

Side-by-side scrollable
presentations

Compare individual text
files or entire directories

Menu-driven interface

Built-in editor allows file
change while viewing the
comparison

Target files can be chosen
from directory comparisons

Ideal for programmers

Requires an IBM compatible with at

least 384K. A hard disk is recommended.

DELTA runs under DOS 2.0 or higher.

Order Now. $79.

VISA or MasterCard accepted.
Also from OPENetwork . . .

The Berkeley Utilities

Enhance your DOS environment with
this set of powerful UNIX tools

Textlib

A screen & keyboard management
library for C developers

Write or call our BBS for information

iNetwork

POWER TOOLS FOR POWER USERS

2 1 5 Berkeley Place, Brooklyn, NewYork 11217
Voice: 7 1 8-638-2240 BBS: 7 1 8-638-2239
1-800-542-0938

Reader Service Number 201
MICRO CORNUCOPIA, #53, May, 1990 69

Build A TTL To Composite
Video Adapter

Have some old composite B&W monitors
you'd like to use? Here's a simple project
that'll give many of them new life.

When you built your clone, did
you wish that you could find a
use for your old composite
video monitor? I had a TTL monitor for
my clone system, but that left me with
nothing to do with my old composite
monitor. I wanted to add the monitor to
a second, 80188-based clone that I had
designed, but it just wouldn't operate at
the IBM frequencies. It wouldn't, that is,
until I built the circuit described here.

Background

Most CP/M systems use a monitor
that operates at a horizontal sweep
frequency of 15,750 Hz and a vertical
sweep frequency of 60 Hz. This, con-
veniently, is the same as standard Ameri-
can television. For a standard monitor,
the sync, blanking, and video all transmit
on a single coaxial cable.

The IBM PC and clones generate a
horizontal frequency of 18 KHz and a
vertical frequency of 50 Hz on a 9-pin
"D" type connector. These frequencies
allow more lines and better resolution,
but they're a problem for standard moni-
tors.

A circuit to convert the TTL-level sig-
nals from a PC or clone is easy. Unfor-
tunately, most composite monitors can-
not display the entire picture; they lose
the first few characters. The problem is
quite simple: the monitors internally
generate a horizontal sync pulse that ef-
fectively blanks the screen for a few mi-
croseconds after the external horizontal
sync pulse ends.

The PC begins sending the first char-
acters almost immediately after the sync
pulse. The monitor ignores these charac-
ters, and the result is a partial display.

Figure 1 — TTL to Composite Video Adapter Schematic

K5YI

+HSYNC

-VSYNC

VIDEO

INTENS

:R5V3

-^ *0*

«"«_&

COMPOSITE
VIDEO OUT

U1 - 74LS123
U2 - 74LS04
U3 - 7406

C1.C2 - MYLAR OR SILVER MICA

70 MICRO CORNUCOPIA, #53, May, 1990

By Stuart R. Bell

741 Okie Ridge
Yukon, OK 73099

So how do you fix this problem? First,
lefs delve a bit deeper into how the
video signal works. A composite video
signal consists of three basic components:
video, vertical sync, and horizontal sync.

The video signal is made up of the
"dots" that the screen displays. The dark
areas of the screen are at the blanking
level. The dots you see are at the white
level.

The vertical sync occurs once per
frame and synchronizes the vertical
sweeping of the screen. The horizontal
sweep occurs once per line and synchro-
nizes the horizontal sweeping of the
screen. We're interested in the horizontal
sync.

On each horizontal line, the PC gener-
ates a horizontal sync pulse thaf s a few
microseconds wide. As mentioned, the
video information begins immediately
after that. The composite monitor cannot
display anything for a few microseconds.

If we could delay the video until the
monitor had recovered from the sync
pulse, everything would be all right. The

circuit to
convert the
TTL-level signals
from a PC or clone
is easy.

first character on the line would appear
at the left side, and everything else on the
line would shift to the right by the same
amount. Unfortunately, this approach
uses a lot of parts to delay the video sig-
nal.

A second approach would be to antic-
ipate the horizontal sync pulse and
generate a new one that would occur a
few microseconds early. This would have
the same effect as delaying the video in-

Figure 2 — Parts List for TTL to Composite Video Adapter

Quan

1
1
1
1
1
1
1
1
1
1
1
1
3
1
1

Description

Comments

74LS123 Ul

74LS04 U2

7406 D3

.01 uf cap Use mylar or silvered mica

.001 uf cap Use mylar or silvered mica

20k, 10 turn pot, Position adjust

500 ohm, 10 turn pot Brightness adjust

100 ohm, 1/4 w resistor

150 ohm, 1/4 w resistor

390 ohm, 1/4 w resistor

470 ohm, 1/4 w resistor

6.8 k 1/4 w resistor

1N914 diode

.01 to . 1 uf bypass capacitors

DB9P connector

Phono jack composite output

Misc. Perfboard, 1C sockets, wire

formation. This circuit works that way,
sort of.

I say sort of because I haven't yet dis-
covered the psychic logic that anticipates
a pulse. Instead, this circuit takes into ac-
count that all the horizontal sync pulses
are identical.

Instead of anticipating a sync pulse
and substituting an earlier one, this cir-
cuit delays the horizontal sync pulse
until it's time to start the next line. Each
horizontal sync is delayed by almost-but-
not-quite a full line. The horizontal
sweep frequency generated by the PC is
about 55 microseconds.

If the horizontal sync pulse for any
given line is delayed by, say, 53 micro-
seconds, and then used to generate the
sync pulse for the next line, the second
line will apparently shift right by 2 mi-
croseconds.

About The Circuit

The circuit (see the schematic in Fig-
ure 1) is straightforward. It uses only
three common TTL ICs. The DB9P con-
nector at the left connects to the circuit
board with a short cable and mates with
the DB9S connector on your video board.
The composite video connector at the
right is a standard phono jack, which
connects to the composite monitor with a
short piece of coaxial cable.

The 500 ohm potentiometer adjusts
the brightness. The 20K potentiometer
adjusts the horizontal position of the pic-
ture by varying the amount of horizontal
sync delay. The circuit generates stand-
ard video levels of V for sync, 0.5 V for
black, and about 2 V for white level.

Building The Circuit

I constructed the prototype on a small
piece of perfboard. The potentiometers
are mounted for easy access with a
screwdriver. I stole 5 V from my power
supply, which is open frame. You could
use an external supply, as long as it

MICRO CORNUCOPIA, #53, May, 1990 71

shared a ground with the circuit board
and the computer.

Be sure to use .01 to .1 pP capacitors
to bypass the circuit. Don't use ceramic
capacitors for CI and C2. If you do, the
picture will probably have the jitters and
will drift. The diode shown in the sche-
matic is a 1N914, but almost any fast
switching diode will work.

Once you've built and connected the
circuit, adjustment is easy. First, adjust
the 500 ohm pot about halfway between
both ends. Then adjust the 20K pot to get
all the characters on the left side of the
screen. If you set the value of the 20K pot
too large, it will not receive sync pulses,
and the picture will be scrambled.

After adjusting the horizontal posi-
tion, set the brightness using the 500 ohm
pot. If you can't get the brightness ad-
justed to your satisfaction, or if the pic-
ture tends to "tear" at the top, increase
the 150 ohm resistor at U3 pin 6 to 220
ohms, and the 470 ohm resistor to 680
ohms.

Adjusting The Monitor

You'll need to adjust most monitors
before you can use them. They usually

have internal oscillators that run at the
horizontal and vertical frequencies of 15
KHz and 60 Hz, and they must be ad-
justed.

The easiest way to adjust the monitor
is to build the circuit, but leave Ul out of
its socket. Jumper Ul pin 1 to Ul pin 5.
This will pass the horizontal sync pulse
through to the monitor. Then connect the
adapter to the video board on the PC and
to the monitor, and set the 500 ohm pot
about halfway.

The vertical adjustment on most
monitors is a pot, and will go down to 50
Hz without any trouble. The horizontal
adjustment may be a pot, and may go up
to 18 KHz.

If the pot doesn't make it (or there
isn't a pot), there's usually a coil with a
ferrite core somewhere near the horizon-
tal pot. There may be another coil that
adjusts the width. Tinker with the hori-
zontal coil and the vertical pot until you
get a stable picture on the screen. Then
install Ul, remove the jumper, and adjust
the adapter.

Cautions

First, be careful working inside the

monitor. The horizontal section of the
monitor generates several thousand volts
for the anode on the CRT. The CRT acts
as a large capacitor and it can hold a
charge for days.

(Editor's note: You'll see a well insulated
wire running from a transformer on the
board to an insulated cap on the side of the
CRT. That's the wire carrying thousands of
volts to the anode, and I'd be very careful
around that entire part of the circuit. Use an
insulated tool when adjusting the core of the
horizontal oscillator, and keep your hands out
of there (even when the monitor's turned off).

Second, keep in mind that some moni-
tors just cannot display the entire picture
of a PC. Most can, but some will lose a
character or two off one end or the other
no matter what you do.

Finally, a (very) few monitors will
overheat if you run them at 18 KHz.
After adjusting the adapter, let the moni-
tor run for a few minutes, then turn it off
and see if anything is excessively hot. If it
hurts to touch it, if s probably too hot.

♦ ♦ ♦

Program FASTER with ...

CC-RIDER

THE C PROGRAMMER'S COMPANION

[NEW!

PROFESSIONAL EDITION

CROSS-REF Editing! Function Prototypes! QuickHelp, too!

Instantly recall your C
symbol definitions by hitting
a hot-key inside your editor!
Along with standard
CCRIDER features like
Source View, Edit and
Definition Paste, the new
Professional Edition has a
Symbol Browse mode and
even lets you walk through
all usages of a symbol. It's
on-line cross-referencing
as you edit!

Build a database of useful
information for your C code
withCCSYM, a powerful
sou/co code analyzer
which can generate fully
commented function
prototype #include files
for your program's static and
global function definitions.
Imagine! - No more manual
maintenance of function
declarations -and you
know they are consistent!

CCSYM can also create a
help database for your
program which is
compatible with Microsoft's
help system, OuickHep. All
symbols in your application
are accessble from
QuickHelp's menus,
together with

documentation extracted
automatically from your
original source code.

Reader Service Number 169

Western Wares

Standard Professional

$89

$279

'?£\;. :,:<3Q3) '327*4893.: V,:
BoxC : NQWQQd >: &) : 81423 :

FotDO$mO$i2

FREE DEMO DISK

BUILDING BLOCKS
FOR THE ASSEMBLY
LANGUAGE PROGRAMMER

QPANTASiffl POWER U!

Price $99.95
With Source Code
$299.95

S/H $5.00

Outside U.S. $15.00

CA Residence

Add 7% Sales Tax

MC and VISA

Over 256 Assembly Language Routines

No Need to Re-Invent the Wheel Every Time

You Write a New Application

Permits Faster Application Development

Overlapping Windowing System in Less Than 3K

Fast and Compact

Ideal For Programming TSR's

C and Pascal Interfaces Available

MASM 5.X Compatible

19855 Srevens Creek Blvd. Suite 154
Cupertino, CA 95014
(408)244-6826

ALSO AVAILABLE:

ASMFL0W$99.95
An Assembly Language
Flow Charting and Source
Code Analysis Tool

Reader Service Number 139

72 MICRO CORNUCOPIA, #53, May, 1990

punaSOG VACATIONFORI99oi

Announcing SOGEast '90

The Eastest SOG of them all returns for an all new show on
August 17 and 18, 1990, in York, Pennsylvania.

This SOG starts with a picnic on the evening of August 16,
around 4 or 5 o'clock. The picnic includes delicious fare fixed
on site for your dining pleasure. There will also be frisbees,
Softball, volleyball, and more. Come in early and spend time
meeting fellow SOGees.

The 17 th and 18 th (Friday and Saturday) will be chock full of
seminars, ranging from Hardware Control Using C++, to DOS
Internals, ONLine Secrets for Modem Lovers, and More. Friday
afternoon will sport a live panel discussing (calmly, I'm sure)
favorite programming editors, including QEdit, MultiEdit, Brief,
and others. If you want to stand up for your favorite, let us
know so we can save you a chair.

Friday evening is the First Annual SOGEast Dinner Party.
This is a sit-down, stand-up, have a ball Ball. Don't change
your clothes from the SOG sessions, just come for the good
time. A special guest speaker will be introduced....

Friday night the infamous Jolt SIG (with real Jolt this year!)
will convene for a night of fun and merriment. A special hard-
ware project especially for software people is scheduled, as
well as a contest for Programmers vying for the title of "The
One Day Wizard of SOGEast." We'll even be awarding prizes!
If you are interested in the project or the contest, let us know.

We are looking for speakers on a large range of topics. If
you'd like to share some knowledge, pass along some secrets,
or amuse us for an hour, please call soon.

Current suggested retail prices are as follows:
Advance SOGEast '90 Registration (before June 1) $25
Advance SOGEast '90 Registration (June 1 - July 15) $35
SOGEast At The Door (Anything after July 1 5) $50

Picnic $10

Official SOGEast '90 Tee shirts $10

Friday Evening Dinner $15

Tables for Exhibitions $25

(All prices are Per Person)
There is more.... For the latest information, contact:

John Ribar

The CDS Group

P.O. Box 7549

York, PA 17404

(717) 792-5108 Evenings

CompuServe 73577,1652

bix jribar

We're proud to announce that SOGEast '91 is already in
planning. This will be an annual event. This year, we plan to
hold the SOG at one of two local colleges. If you have sugges-
tions for next year, pass them on.

<><-sZ:?<^»^

MICRO C ORN ucoPM,«, May , I99073

continuing AROUND THE BEND

Micro C Garage Sale

u^ ed

Ssx?&

s PecL

ial

130

900

cl °seout!

While Quantities Last!
No rainchecks!

Genuine Antiques!

No Cliches!

All Sales Final!

Visual 50 Terminal (VT-52 compatible) like new $60

Boxes of new 8" disks, 10 disks per box, SSSD $5/bx

Commodore 128 w1571 floppy drive, used 4 hrs. $100

8 1/2" reals of 1/2" mag tape, unopened $10 ea

8" SS SD/DD Floppy Drives, Mostly Shugart 801s $10 ea
Cabinets, new, can hold 2 8" drives, includes PS $10 ea
Complete Cromemco S-100 sys. w/boot disks, manuals $75
Kaypro II-83, very used, working $100

Kaypro 4-84, very used, working $100

Big Board I with cabinet, two 8" drives & PS $75

Big Board I in keyboard case with PS $50

Baby Blue Z80 cards, run in 4.77 MHz XTs, w/utilities $10
Constant voltage transformers, Sola, 250 V.A. 1 10V $35 ea
2764s, Intel 200 ns, new (in tubes of 14) $21/tube

Call for other goodies: For instance we have back issues of Life-
lines (Lifelines?) and Computer Journal (CP/M anyone?). Plus, we
have numerous computer books.

All items sold as is. No refunds. All worked the last time we used
them (even the books). Call for availability as well as shipping
and handling charges. Shipping will be significant on some items
so if you plan to be in the Bend area, definitely give us a shout.
(Shipping and handling are $7.50 (in the U.S.) for the Baby Blue
Z80 cards and $3.00 per order on the ROMs.)

To Order, Call Today, 503-382-5060

(9 a.m. to 1 p.m. Pacific Time)

puter magazines (or how far they'd moved away from us).
What's important about CL's subscription guarantee isn't that
the magazine might replace Micro C, but that you can, without
risk, try it. You can see if it, too, doesn't become a valuable
resource.

I must also mention a couple of other magazines:

Programmers Journal (503-747-0800) is a smaller, less formal,
version of CL. It wouldn't surprise me to see PJ really grow,
both editorially and in circulation over the next few years.

Midnight Engineering (303-225-0856) could become another
Micro C. Bill Gates is aiming ME at the folks who are working
graveyard "On Their Own." Several Micro C regulars had ar-
ticles in his first issue.

Meanwhile, if you want CL to look more like Micro C,
you'll just have to write articles for them. Contact Michelle
Williams at (415) 397-1881 for writers' guidelines.

We've already given the articles we've been working on to
J.D. at Computer Language and referred several embedded
authors to Tyler Sperry (a very old friend) at Embedded Sys-
tems.

Bruce Eckel's already writing for J.D. Karl Lunt, Bob Nan-
sel, John Ribar, and Michael Hunt have already connected
with Tyler. Who knows, you might even see the Fogg in one
or both.

The Office Will Be Open

Starting April 2, we'll have a skeleton crew (Nancy) man-
ning (womanning?) the office from 9 a.m. to 1 p.m. Pacific
Time. Nancy' 11 keep the office open at least through April,
probably through May. Sandy and I will no doubt be hanging
around the office, too.

We'll still take orders for back issues, XT and Kaypro sche-
matics, as well as issue disks. Speaking of back issues, we've
hired a professional indexer to thrash through issues 1
through 52. It'll be on disk (on paper, too, we hope). We'll
send you the index for only $6. Or you get the index disk free
with an order for 10 or more back issues. It's best to order
soon; we're getting low on quite a few issues.

If you're into bargains (I'm talking historical treasures
here), check out the Micro C garage sale ad on this page. We
have lots of new 8" SSSD disks, 10 to a box (make great bases
for little league). We have 8" single-sided drives (take a rope,
tie a drive on one end, tie a boat on the other....).

We have (shudder) Big Boards for sale, a terminal, constant
voltage transformers for those of you with dirty power, and
something very, very special. We've got about 70 Z80 plug-in
cards. I was hoping to announce them at the end of the
LIMBO series (but we ended before the series did). You could
use them to develop ROMs for the 64180 (or run any other
CP/M software on your PC/XT/clone). One problem, they
don't seem to work on any system running faster than 4.77
MHz. However, they're only $10 each (including software)
plus $7.50 each, shipping and handling.

It's best to order by phone (503-382-5060) so you can be
sure we've got what you want. (Or that you want what we've
got.)

74 MICRO CORNUCOPIA, #53, May, 1990

AROUND THE BEND

Where Am I Headed?

Nowhere.

I want to have some time that isn't already spoken for. I've
spent my whole life in student mode. You know, if I'm awake
and not working, then I should be.

I can't think of anything more enticing right now than
spending two weeks or two months on a sailboat anchored off
a deserted island. By the end of two months, I'd be totally stir
crazy. I hope.

Finally, I've been feeling at odds with everything lately. I've
been tense, uncomfortable, absent-minded, jumpy, irritable —
all reminding me that Micro C and I are very close. But this
issue's edit — struggling line by line through every article — has
been nearly impossible. Each evening I take home the articles
which need my touch. Each morning they return untouched.

My heart's not in it.

Sorry.

My Secret Wish

People who've called after hearing rumors of the end have
asked how I'm feeling ("sad" is the best I've come up with)
and what I'm doing next (nothing). No one has asked if some-
thing has been left out — if there's been something I've hoped
would happen, but hasn't.

Yes, there is something. For most of these nine years, I've
secretly wished that just once, for a day or an hour, every sub-
scriber, every newsstand reader, every library browser who
regularly thumbs through Micro C, every one of you who has
followed and shared this journey — could get together. All of
us in one place at one time. Twenty-five thousand incredible
people.

That was my wish.

Solid Expectations

As of January 1, Miniscribe Corporation is under court pro-
tection from its creditors. The company threw in the brick after
a year of shipping half-baked clay blocks from warehouse to
warehouse, pretending they were hard drives.

I'm sure the bankruptcy surprised folks from the building
trades; after all, the demand for good, new bricks has never
been higher. However, these small red clay thingies aren't
commonly used to hold data.

That may change shortly. We've received late word (very
late word, actually) that Taiwanese computer manufacturers
are scrambling to retool their 256K production lines. Ap-
parently there's a rumor running rampant around the island
that bricks are the latest in solid storage technology.

Short Note

The biggest Taiwanese exporter is IBM. (Wait until they
hear about bricks.)

More Bricks

I was talking to Allyn Franklin, my favorite hard drive per-
son (who knows, he may also repair bricks), about his "On
Your Own" article when he offered me a few tips on dealing
with a recalcitrant hard drive.

Develop Powerful

Image-based Applications

Quickly and Easily

Introducing VICTOR, video capture
and image processing library

Create image-based applications that display live video on
the VGA and offer powerful image processing features

With Victor your applications can show live video on any portion of the VGA
screen and combine live video with text and graphics. Use sophisticated im-
age processing tools, create lineart, TIFF/PCX files,
more. Victor handles all the low level functions so
you can concentrate on your application.

Rapidly test and prototype

Victor comes with ZIP Image Processing soft-
ware to get you up and running quickly - test and
prototype functions before you write any code.

Video frame grabber support

Victor supports EGA and VGA up to 800 x 600,
and video digitizers by: IDEC, CCI, Micromint,
DAK, HRT, Electrim, Willow, VG, and others.

Victor supports Microsoft C, QuickC, and
Turbo C, includes demonstration and prototyp-
ing software, full documentation, and source code
for device control routines ... all for only $195.
Victor with video frame grabber $349

Call (314) 962-7833 to order

VISA/MC/COD

Image-based applications can
be developed in QuickC and
Turbo C environments.

Your applications will have
powerful image processing and
handling capabilities.

Catenary Systems 470 Belleview St Louis MO 631 19 (314) 962-7833
Reader Service Number 217

on NEW PLATFORMS

VAX/VMS

HP 9000/300

XENIX/386

Motorola UNIX V.3

Sun-3

NCR Tower

Motorola UNIX V.3

VAX/VMS &ULTRIX

Sun-3, 386i

XENIX/386

OREGON C++

■ A true compiler - not a translator

■ Three compilers in one: C++, ANSI C or K&R C

■ Source-level debugger

■ Conforms to AT&T V2.0 C++

PASCAL-2

■ Full ISO Level 1 standard

■ Comprehensive error checking

■ Access to command-line arguments

■ Extensions include full string type, separate

compilations, external variables & more

XENIX/386

Sun-3

VAX/VMS

OREGON MODULA-2

■ Conforms to emerging ISO standard

■ Excellent support of coroutine and interrupts

■ Ideal for embedded systems

680X0 CROSS-COMPILERS ALSO AVAILABLE

CALL 1-800-874-8501

(503) 245-2202 FAX: (503) 245-8449

OREGON fjfe SOFTWARE

6915 SW Macadam, Suite 200, Portland, OR 97219 USA

The following are trademarks: Oregon Software, Oregon C++, Oregon Modula-2, Pascal-2, Oregon
Software, Inc., Xenix, Sun-3 , 386i , VAX, VMS, Motorola, UNIX, AT&T, ULTRIX, NCR Tower, HP 9000

Reader Service Number 193

MICRO CORNUCOPIA, #53, May, 1990 75

AROUND THE BEND

BIOS

SOURCE
CODE

The AT BiosKit gives you a complete Bios with source
code you can modify for your own applications! The
BiosKit includes a Bios on diskette ready for program-
ming an Eprom, and includes the utilities you need to
Rom the source code. The Bios also has a Rom
Monitor/Debug and Setup. At last you have control over
the core of your system. Over 380 pages, with diskette,
$199. The XT BiosKit is only $99, or get both for $279.
The Intel Wildcard Supplement for the XT BiosKit is $49.
We'll include a free copy of the
pocket-sized XT-AT Handbook by
Choisser and Foster with each BiosKit if you men-
tion this ad when you order. Of course, this $9.95
value is also available by itself. Or buy five or more for
only $5.00 each.

FREE

MasterCard]

VtSA-

800-462-1042

In California 619-271-9526

Annabooks

12145 Alta Carmel Ct Suite 250-262

San Diego, California 921 28 Money-back guarantee

Reader Service Number 160

fifilxf

It's Glorious!

"No color PCompatible is complete without a copy of this;
get one and see what I mean." - Jerry Pournelle, BYTE

"Infinite... Neither of us (the chief engineer of the Galileo
Jupiter probe and I) may ever be heard from again."

-Arthur C.Clarke

"I could watch an animated image for hours."

Larry Fogg, Micro C (MIA 89)

FractalMagic 5.0

*" America's Premier Fractal Program

IBM PC, PS/2 (3.5" or 5.25")

CGA, EGA, & VGA: $35.00

Macintosh: Plus, 512KE, SE,

SE/30, and Mac ITs: $35.00

IBM Expansion Modules 1-4:

$15.00 each, or all four for $45.00

Sintar Software

"Software for the Mind"

1001 4th Avenue, Suite 3200

Seattle, WA 98154

(206) 625 - 1213

Visa - MC - AmEx

r™^" -H.^-*-™—--

Qg»(cuti»»

"If something seems weird, don't turn the drive off. You
may have a damaged boot sector. Just back up everything you
hold near and dear.

"Often, however, you don't know there's been a problem
until you power up the next day. And, what often happens is
the system doesn't realize it has a hard drive. That's when I
run FDISK. Try all the versions of FDISK you've got. FDISK
might see the drive." (I just tried it. It works!)

"If you run FDISK with the same parameters as you did
during the original format, it won't eat any data. Normally
you have to do a high level format after FDISK, but in this
case you want to avoid that. With any luck, after you've run
FDISK, the system will again see your drive."

"If FDISK doesn't help, try reformatting just track 0. Mace's
FORMAT-H is the tool I use a lot for this.

"You can reach Western Digital at (800) 832-4778. Key in
the type of drive and WD controller number and the system
will tell you how to jumper the card, and will give you head
and cylinder counts for the drive.

"There's an even longer list inside the newest Speedstor,
and Tandon has recently come out with DRIVES.COM.
DRIVES.COM queries your AT and your hard drive, telling
you which drive number to enter, etc.

"Both Mace and Norton have books out on data recovery,
but neither goes far enough. They are really aiming at the
average user."

Drives That Should Have Been Mortared

My brother grabbed me the other day, desperately looking
for a letter we'd run. He wanted the issue and page number of
the bit about 15 Seagate 238s that died. He wanted the copy
for a customer who wouldn't accept an RLL drive (and didn't
much care for MFM either).

Don had asked the guy if he was running 238s and 225s.
Turned out he was. And they'd failed.

If I were buying a hard drive right now, I'd get: A Seagate
full height like the 8096, 8048 or 8038; a Miniscribe full-height;
or, even better, a Mitsubishi. The Mitsubishi I'm thinking of
can run either 40 meg MFM or 60 meg RLL. If I were con-
cerned about speed and space, I'd run the Mitsu RLL. On the
other hand, there's nothing safer than an RLL certified platter
plodding along with MFM data.

76 MICRO CORNUCOPIA, #53, May, 1990

Micro Cornucopia

Back Issues

Final Issue

! Stock Up

on Back Issues Now!

ISSUE #1(8/81)

ISSUE #17 (4/84)

ISSUE #30 (6/86)

Accessing dBase III Plus

Kernel

Power Supply

Voice Synthesizer

PROLOG On The PC

Records from Turbo Pascal

The Very Early Days of

1/2 PFM.PRN

68000-Based System Review

Expert Systems

96 pages

Computing

16 pages

56 pages

Logic Programming
Building Your Own Logic

ISSUE#44 (11/88)

96 pages

ISSUE #2 (10/81)

ISSUE #18 (6/84)

Analyzer

Object-Oriented Program-

ISSUE #49 (9/89)

Parallel Print Driver

Kaypro EPROM Programmer

PC-DOS For Non-Clones

ming

I/O, I/O...

Drive Motor Control

I/O Byte: A Primer

104 pages

A Taste of Smalltalk

Build A Computer The Easy

16 pages

Serial To Parallel Interface

Actor

Way

Business COBOL

ISSUE #31 (8/86)

Thinking Objectively

PostScriptals

ISSUE #3 (12/81)

60 pages

RAM Resident PC Speedup

Building MicroCad

Prog. Logic Controllers

Configuring Modem 7

Practical Programming In

Peripheral Technology-

Driving Stepper Motors

Reverse Video Cursor

ISSUE #19 (8/84)

Modula-2

PT68K-2

Writing TSR Programs

FORTHwords Begins

Adding Winchester To BBII

Game Theory In PROLOG, C

96 pages

Low Cost I/O For The PC

16 pages

6 MHz On The BBI

104 pages

Interfacing 1 6-Bit Devices

Bulletin Boards

ISSUE #45 (1/89)

96 pages

ISSUE #4 (2/82)

Track Buffering On Slicer

ISSUE #32 (10/86)

Computer Aided Design

Keyboard Translation

4 MHz For The 820- 1

Public Domain 32000:

CAD In A Consulting Business

ISSUE* 50 (11/89)

Modems, Lync, and S10s

64 pages

Hardware And Software

Choosing PCB Layout Systems

3-D Graphics

Undoing CP/M ERASE

Writing A Printer Driver for

Building Circuits With Your

3D Surface Generation

20 pages

ISSUE #20 (10/84)

MS-DOS

Computer

PC Video Frame Grabber

HSC 68000 Co-Processor

Recover A Directory By

Secrets of Optimization

UMBO, Part Three

ISSUE #5 (4/82)

DynaDisk For The BBII

Reading & Writing Disk

Finding Bargains in the

PostScript, Part Two

Two Text Editors

Serial Printer On BBI Sans S10

Sectors

Surplus Market

UNIX For The PC

Double Density Review

Cheap & Dirty Talker For Kaypro

96 pages

MASM 5.1

Capturing & Graphing A Voice

20 pages

72 pages

ISSUE #33 (12/86)

96 pages

In Real Time: Part 1
96 pages

ISSUE #6 (6/82)

ISSUE #21 (12/84)

ISSUE #34 (2/87)

ISSUE #46 (3/89)

BBI EPROM Programmer

Analog To Digital Interface

ISSUE #35 (4/87)

Software Tools

ISSUE #51 (1/90)

Customize Your Chars

Installing Turbo Pascal

ISSUE #36 (6/87)

The Art of Disassembly

Embedded Systems

Double Density Update

Low Intensity BBI Video

ISSUE #37 (9/87)

Handling Interrupts With Any C

Embedding An XT Motherboard

24 pages

80 pages

SOLD OUT

Hacking Sprint: Creating
Display Drivers

Writing A Neural Network in C
LIMBO, Part 4

ISSUE #7 (8/82)

ISSUE #22 (2/85)

ISSUE #38 (11/87)

Greatest C Compilers

Getting Started in Hardware

6 Reviews Of C

Xerox 820-II To A Kaypro-8

Parallel Processing

Turning A PC into An

Capture & Graph a Voice,

Adding 6K Of RAM

Sound Generator For the

Laser Printers, Typesetters

Embedded Control System

3D Surface Generation, Pt 2

On Your Own Begins

STD Bus

Build A Graphics Scanner

96 pages

Marketing Your Own Software

24 pages

Reviews Of 256K

For $6, Part 2

96 pages

RAM Expansion

Writing A Resident Program

ISSUE #47 (5/89)

ISSUE #8 (10/82)

88 pages

Extractor In C

Robotics

ISSUE #52 (3/90)

SOLD OUT

96 pages

The LIMBO Project

C++ 2.0

ISSUE #23 (4/85)

Starting A Robotics Company

Training A Neural Network

ISSUE #9 (12/82)

Automatic Disk Relogging

ISSUE #39 (1/88)

How To Write and Use A

Debugging C Pointers Using

BBII EPROM Program

Interrupt Drive Serial Printer

PC Graphics

SystemProfiler

MEM

Relocating Your CP/M

Smart Video Controller

Drwing Mandelbrot / Julia Sets Problem Solving and Creativity

The AT Keyboard Interface

Serial Print Driver

Review: Microsphere RAM Disk

Desktop Graphics

Turn Your XT Into A Controller

Filling In The Holes On Your XT

Big Board I Fixes

86 pages

Designing A PC Work-

Writing Code For Two

Logic Families

32 pages

station Board

Operating Systems

EPROM Programming

ISSUE #24 (6/85)

Around the TMS-34010

96 pages

Controlling Runaway C With

ISSUE #10 (2/83)

C'ing Into Turbo Pascal

96 pages

CC- Rider

ISSUE #11 (4/83)

8" Drives On The Kaypro

ISSUE #48 (7/89)

96 pages

SOLD OUT

68000 Versus 80x86

ISSUE #40 (3/88)

Tools For The Physically

Soldering: The First Steps

The Great C Issue

Impaired

♦ ♦♦

ISSUE #12 (6/83)

88 pages

11 C Compilers

The Adventure Begins

Bringing Up BBII

Writing A Simple Parser In C

Selecting A Talking Computer

Double Sided Drives for BBI

ISSUE #25 (8/85)

C++, An Object Oriented C

For A Blind Friend

^-^Tts.

Packet Radio

Why I Wrote A Debugger

Source Level Debugger For Writing Software For The Blind

>g|^

5 MHz for Kaypro

The 32-Bit Super Chips

Turbo C

File Transfer Via Parallel Port

46 pages

Program The 32032 Modula II

96 pages

The LIMBO Project— Part Two

fiK§Pllb

RS-232C: The Interface
1 04 pages

ISSUE #41 (5/88)

PCX Compatibility

A 68000-Based Multitasking

»B5s3^— — =-*T

ISSUE #13 (8/83)

iiB\r —i 3

CP/M Disk Directory

Artificial Intelligence

^P™^» '

More 256K for BBI

ISSUE #26 (10/85)

3-D Graphics

"^^ZtcLxL

Cheap Fast Modem
BBI Printer Interface
44 pages

Inside ZCPR3

Two Megabytes On DSI-32

SOGIV

Neural Networks

Logic Of Programming
Languages

To Order:

The Future Of Computing

Applying Information Theory

*^^^^^^™

ISSUE #14 (10/83)

Graphics In Turbo Pascal

96 pages

Phone: 1-503-382-5O6O

BBII Installation

104 pages

Mail: PO Box 223

The Perfect Terminal

BBI Video Size

ISSUE #27 (12/85)

ISSUE # 42 (7/88)

Bend, Oregon 97709

Video Jitter Fix

SOLD OUT

Maintaining PCs

Keeping Your Hard Drives

48 pages

United States,

ISSUE #15 (12/83)

ISSUE #28 (2/86)

Rescuing Lost Text From

Running
Troubleshooting PCs

Issues #1-34

$3.00 each ppd.

Screen Dump Listing
Fixing Serial Ports

Memory

Introduction To Modula-2

XT Theory of Operation
Simulating A Bus

Issues #35-current

$3.95 each ppd.

Playing Adventure
48 pages

Inside The PC

96 pages

104 pages

Canada & Mexico

ISSUE #16 (2/84)

ISSUE #29 (4/86)

ISSUE #43 (9/88)
Building Databases

All issues

$5.00 each ppd.

Xerox 820 Column Restarts

Speeding Up Your XT

Build a C Database

BBI Double Density

Prototyping In C

Selecting a dBase III

Foreign (air mail)

BBII 578" Interface Fix

C Interpreters Reviewed

Compatible Compiler

Recovering Text From Memory

Benchmarking The PCs

Working with Paradox

All Issues

$7.00 each ppd.

52 pages

104 pages

Designing Custom PC Cards

MICRO CORNUCOPIA, #53, May, 1990

Tired of the limitations of MS-DOS ?
Thinking about switching to another OS ?
Weil DONT, now there is :

CLI is an intelligent front & back end processor
that enhances the MS-DOS system kernel. CLI
provides many features of advanced OS's, like
VMS and UNIX, to MS-DOS without any re- training.

The core of the CLI kernel is based on recursive
definitions and references to Alias, Logicals,
Symbols, and Lexical-Functions from the command
line, batch programs, and existing application
programs (without any modifications).

Powerful command programs can be quickly
written for a wide range of purposes such as:

• flexible control of code & data management

• code development & code support tools

• automated file/device malntanence utilities

• creation of generic command procedures

• installations & system re-configurations

For the first time, designers, developers, and
managers can coordinate software development and
production efforts without the major limitations
and restrictions in the MS-DOS environment.

• VMS is a trademark of D.E.C.

• ONIX is a trademark of AT&T

• MS-DOS is a trademark of Microsoft

1005 Sussex Blvd
Broomall, PA 19008

Price: $149

S/H: $ 6

215-543-2658

Reader Service Number 187

16 MHZ
68000

SINGLE BOARD COMPUTER

The new PT68K4 computer board has all the features of the K2 series
plus extra benefits. It is available in 2 speeds - 12MHZ and a super fast
with increased memory capacity to a maximum of 12 MB!

• 4 RS232 Ports • 7 XT Expansion Slots

• Floppy Disk Controller • 1 Memory Expansion Slot

• Real-Time Clock • 2 Parallel Ports

• Optional Operating Systems

BASIC KIT (12MHZ) - 4 layer board, Processor, Humbug Monitor,

4K SRAM, 2 Serial Ports $220

K4 COMPLETE KIT (16MHZ) - with 512K of DRAM $545

ASSEMBLED BOARD (16MHZ) - with 1MB DRAM $649

PROFESSIONAL 0S9 with C COMPILER $299

only valid if purchased with board

SK*D0S OPERATING SYSTEM: $80

COMPLETE INFORMATION AVAILABLE UPON REQUEST

PERIPHERAL TECHNOLOGY

1710 CUMBERLAND POINT DR. #8

MARIETTA, GA 30067

404/984-0742 *FAX 404/984-8248

COD/MASTERCARD/VISA/CHECK

SK*D0S IS A TRADEMARK OF STAR-K SOFTWARE SYSTEMS
0S9 IS A TRADEMARK OF MICR0WARE AND MOTOROLA

Reader Service Number 119
78 MICRO CORNUCOPIA, #53, May, 1990

AROUND THE BEND

NEC Can't Defuse Printer Problem

It's 3:57 p.m. It's quiet in Carol's office— her NEC 890 Silent
Writer is unusually silent.

"Dave, my printer doesn't work."

The power, which had been off for an hour, had just come
back on and all the computers were doing their deep knee
bends.

"Dave, my printer doesn't work."

"So give it a chance, already."

Unfortunately, it needed more than a chance. Nothing,
dead, no fan, no lights, no smoke, no whimper, no nothing. I
looked for a fuse. None in sight. None underneath, on top, by
the plug, anywhere. I opened the back and pulled out the
power supply. No fuse.

I called NEC. After being referred to three different sites, I
finally reached someone who would talk to me....

"Where's the fuse on the 890 Silentwriter?"

"Before we talk about that, what are the symptoms?"

"It's dead."

"It could be something other than a fuse."

"Right, but for now I'd really like to know where you put
the silly fuse."

"I'm not sure. Have you tried to reboot it?"

"Reboot it? REBOOT IT?"

Of course, what was I thinking. Of course, I'll need an
operating system for a dedicated 68000. Let's see, DOS 3.1
doesn't run very well on an 8088.

Now, without a floppy drive, I'm not sure where to put the
3.1 boot disk (and I'm certainly not open to suggestions). And
with a totally dead system, how will I know when the silly
thing's been booted sufficiently? (Don't say it, I'd just hurt my
foot.)

800 Frustrations

I can name two significant events in the financial life of
Micro C. First, we got okayed as a VISA /Mastercard mer-
chant, then we got an 800 number.

We had a dickens of a time becoming a VISA merchant. I
was talking to the last bank in town when I finally got the
okay. Banks seem awfully gun shy about small mail order
businesses like Micro C. Once we'd been okayed by one bank,
however, everyone wanted to step in. (No way.)

Working with VISA has been interesting. Wives cancel
husbands' charges. Card numbers that are okay one day are
not okay the next. People use other people's charge accounts.
But by and large, VISA has been great.

Getting an 800 number was a lot easier but a lot more ex-
pensive. We pay a monthly service charge plus a per-minute
fee during calls. The fee is often higher than the normal direct
dial rate, and like standard long distance, we pay more for
east coast calls than for west coast. But, an 800 number makes
it possible for people to order subscriptions or disks from
work, from a friend's house, or anywhere.

But I was worried that "orders" might become half-hour
tech calls. What I didn't worry about was misdialed calls. I
should have.

None of the 800 services tell you they assume no responsi-
bility for wrong numbers. Initially that wasn't a problem. We
received maybe two or three erroneous calls a week. Lately,
however, we've gotten as many as 15 wrong numbers per day

AROUND THE BEND

(besides our usual half-dozen calls for subscriptions).

A travel club called Excellence in Exercise is advertising
tours to east coasters. Their number is 800-888-TOUR. Great.
That's 800-888-8687. But if people dial zero instead of the O in
TOUR, they get 800-888-8087. That's Micro C.

Calling the company hasn't helped, after all they're doing
nothing illegal. So don't be surprised if you dial our 800 num-
ber after April 1 and you hear it's been disconnected. You can
still use our regular order number: 503-382-5060.

(But maybe we're missing a much better solution. We could
just put out the word there's a new toll-free Lotus 3.0 support
number — something ending with 8687.)

Who's Selling Computers?

I've been watching The Wall Street Journal again. IBM, Apple,
and Radio Shack are all reporting that computer demand is
drying up. Meanwhile, Microsphere is scrambling to keep up
with orders.

I called my brother.

"Hello, is Don there?"

"I'm sorry, he's tied up with a customer."

"But I'm his brother."

"Then he's especially tied up."

Busy is one thing, but being tied up with a customer is
pretty kinky.

Later, after things unraveled a bit, I asked him what was
happening in the industry.

"You know, Intel's profits are way up. It's all those 386s
they're shipping. People are buying 286 and 386 systems like
they're going out of style.

"Meanwhile, Radio Shack is now suffering from its reputa-
tion for incompatibility. They set up their systems so that you
have to buy their drives, their memory cards, their everything.

"And, Apple has finally dropped its low-performance pro-
ducts. The variations on the Apple II, for instance, were cash
cows for Apple. Now they're gone. Plus, the older MACs are
selling just as slowly as they run.

"Even some of the big clone houses have set up their house
brands to require custom memory boards. Really helps their
bottom lines — for a while."

So it appears that the demise of the computer industry is
really more of the decline of its figureheads.

PostScript

There's nothing more worthless than a PostScript printer
when you want to output regular stuff. So you connect a silly
dot matrix printer to your PostScript machine. Nothing wrong
with that except dot matrix is slow, noisy, and its output isn't
very clean looking.

When I heard that Legend had come up with a simple TSR
package that automatically grabs anything sent to, say, LPT2:
and sends it out, say, LPT1: wrapped inside PostScript, I called
them up.

I haven't tried all the graphics or font options available on
an Epson FX-85 or IBM graphics printer, but what I have tried
has worked. Painlessly. You can run it as a TSR or as a filter.
Either way it's a godsend for outputting the ordinary stuff on a
super printer, super fast.

, " ROBOTIC

U SYSTEMS

LIMBO Parts Kit Series

Issue #50 Mobility Base

Issue #5 1 Superstructure
& Controllers

$219
$289

Please allow 3-4 weeks for delivery. Prices
subject to change without notice. International
orders must be paid in advance in US funds
drawn on a US bank. Visa, Mastercard, personal
check, and COD accepted. Shipping &
Handling, $10. Wisconsin residents please add
sales tax. Call or write for more information.

(414j 769-9070

3620 East Layton Ave, Suite 25 ■ Cudahy, WI 53 1 1 0-1 462

Reader Service Number 166

Turbo Pascal Programmers... are scraps
of paper keeping you awake nights?

Tired of digging through
old notes? Can't re-
member whether a func-
tion returns a real or an
Integer? Want to get
procedure calls right the
first time?

Then you need MyFLIN.

• It's a TSR. Hit the hot key to display procedure Information
right beside the code you're editing.

• It's easy to find Information. Type part or all of a procedure
or function name, and press the hot key MyFlin will pop
up with the nearest match(es).

• To save new procedure Information.Just place your cursor
on the same line as the procedure, pop up MyFLIN. and
tell it to add the procedure to the database. Requires
PC/XT/AT Computer. MS-PC -DOS 2/3.xx. That's It.

Throw away the paper, get MyFLIN and get going.

OpalFire.

Software Inc.

329 North State Steet,
OREM, UTAH 84057
Phone 1-800-336-6644

Price $30.00 ppd

Visa/Mastercard/American Express

Reader Service Number 161

MICRO CORNUCOPIA, #53, May, 1990 79

YOU WANT THE SOURCE ?!

WELL NOW YOD CAM HAVE IT! The MASTERFUL
DISASSEMBLER (MD86) will create MASM compatible
source code from program files (EXE or COM).
And the files are labeled and commented so
they become USEABLE. MD86 is an interactive
disassembler with an easy to use,, word
processor like interface (this is crucial for
the REAL programs you want to disassemble).
With its built-in help screens you
won't have to constantly refer to
the manual either (although
there are valuable discus-
sions on the ins and outs
of disassembling which
you won't want to miss).

M086 is a professionally

supported product and yet costs

no more than "shareware". And of course, it's

not copy protected. VERSIOH 2 KM AVAILABLE!

MD86 ¥2 is OMLT $67.50 ($1.50 sth) + tax

C.C. Software, 1907 Alvarado Ave., Walnut
Creek, CA 94596, (415) 939-8153

Reader Service Number 31

circuit emulate- c Pas c»l. B«- «

lM mp-ete" sWcuse '' S

, math eopn>™««

emulation sup°°«

.as-* -»<

. ; M ,e*i«i«i» rt , Ma „ roW emv.itnomTu.bo

rr-* iw :: 2 9 5 oo

B , x . -pin paradox > ed

Visa/Mastercard'CO"

^W,

APADIOJV]

Reader Service Number 113
80 MICRO CORNUCOPIA, #53, May, 1990

AROUND THE BEND

PSFX $85

Legend Communications

54 Rosedale Avenue West

Brampton, Ontario

Canada L6X 1K1

(800) 668-7077

Thanks

I can't finish without thanking a whole lot of you.

Thanks to all the folks who attended our local SOGs (espe-
cially to Andy Bakkers from Holland, who attended every
one). You made those SOGs very special for me.

Thanks to all who organized, staffed, and attended the re-
gional SOGs: including a special thanks to Dave, Randy, Mike,
Karl, Bob, Kim, Don, and John — those of you who did it, and
you who are crazy enough to do it again.

Also, thanks to all who subscribe. Many publications spend
most of their money and time selling — going door to door to
find new advertisers and mailing zillions of promotional
pieces to find new subscribers. Though we haven't done any
circulation promotion for TS/i years, we're very close to our
peak circulation. That doesn't happen very often in this in-
dustry.

The reason we've done so well is you. You've stayed with
us. You've encouraged others to subscribe. In addition, you've
been willing to read the world's longest editorials year after
year after year without complaint. (Okay, you caught me. But
three of you haven't complained.)

I'd like to thank my mother and father, without whom I
wouldn't have parents. And, I'd like to thank my father-in-
law, who was one of Micro C's biggest supporters and a
wonderful friend.

I'd like to thank Jennifer and Erin. (They're like family to
me.) It's one thing to have a job, if s another to conceive and
raise a magazine. Over the past nine years, they've played sec-
ond fiddle (and second piano) to 53 editorials, 53 deadlines,
and 53 hassles with the printer. Now I'm going to have some
time, and they're going to have a Dad. (They may insist I start
another magazine, real soon now.)

Plus there's Sandy. If everyone had a Sandy, there'd be a lot
more successful businesses. I'm not sure which of us has
worked harder to make Micro C go, and I'm not sure which of
us is more burned out.

Finally, there's the staff. What can I say about this collection
of miscreants who — put up with, cover for, tolerate, even oc-
casionally get a laugh out of — what's left of me. They've pro-
duced magazines, harassed writers, filled pages with advertis-
ing, and still found time to party. A truly great group.

If You're Bummed

I'm sure many of you won't appreciate this change. I under-
stand. It's part of the sadness I feel. If you feel strongly about
it, write to me, call me, or leave a message on the BBS. I'll be
here.

For a while.

David Thompson
Ex-Editor & Publisher

Shortest Distance

UNITS &
MODULES

By Michael Hunt

2313 N. 20 th
Boise, ID 83702
(208) 336-7413

Michael concludes his column with a look
at connections: tying together network
nodes, for instance, with a minimum amount
of cable. An interesting problem.

Everyday we use our reasoning
ability to find the shortest distance
between two points. The trip to
work, to the store — you know how
routine it is. But how good are we at
finding the shortest distance between
many points? If you've ever set up a
computer network, you were probably
concerned with keeping cable cost down.
The nodes in the network are the many
points that we need to connect.

For a small network, the solution can
be simple and straightforward. As the
number of nodes increases or offsite lo-
cations are introduced, the problem of
efficient connectivity grows quickly. We
need an algorithm to find the shortest or
lowest cost method of connecting all the
nodes in the network.

Abstraction

You can solve the network problem
by using a mathematical structure
known as a graph. A graph consists of a
set of points or vertices, and the lines
connecting the vertices, called edges. The
nodes of the network can be viewed as
the points or vertices of the graph. The
edges represent the connections between
the nodes. You assign a cost to the edge
or connection between two nodes, repre-
senting the distance between the two
nodes.

The graph is a collection of edges,
each with a cost, between vertices. The
set of edges is called a spanning tree. The
sum of the costs of the edges in the tree
is the cost of the spanning tree. Our goal

is to find a minimum cost spanning tree.

Two algorithms address this problem.
The first is known as Prim's algorithm.
Prim's begins with one vertex and builds
the spanning tree one edge at a time. It
begins with a set of one vertex.

For each iteration, it finds the shortest
edge to a new vertex and adds the vertex
to the initial set until all vertices have
been included. Prim's uses an adjacency
matrix to store the edges and their costs.

The other algorithm is Kruskal's. I've
chosen to implement it because it uses a
linked list of the vertices and edge sets.
The linked list is dynamic, not restricted
by some upper array limit like the adja-
cency matrix.

If there are many fewer edges than
the square of the number of vertices,
then Kruskal's algorithm is best. But if
the number of edges is about the same as
the number of vertices squared, Prim's
performs better.

Kruskal's begins with a graph con-
sisting of all the vertices and none of the
edges. Each vertex is in a connected com-
ponent by itself. The algorithm builds a
collection of connected components, each
component consisting of a set of vertices
and the edges that form its spanning
tree.

The list of edges is sorted by cost and
then examined in order of increasing
cost. If the edge connects two vertices in
two different components, the edge is
added to the spanning tree and the two
components are combined. If the edge
connects two vertices in the same com-
ponent, it is discarded. When all the ver-
tices are in one component, the spanning
tree is of minimal cost.

'HC705 Simulator

The simulator is complete, and by
press time Karl Lunt and John Ribar will

e need an
algorithm to find the
shortest or lowest
cost method of
connecting all the
nodes in the
network.

be testing it. John is finishing up the
back end of the compiler. Look for it in
Embedded Systems Programming. For you
Amiga fans, a friend of mine is porting
the simulator to the Amiga using Bench-
mark Modula-2.

Last Column

Well, after nine issues, this is the last
installment of the Units & Modules
column. After talking with Karl Lunt at
Rocky Mountain SOG, I have been bitten
by the hardware bug. Recently I've been
gearing up for the change. Let's see,
multimeter, breadboard, power supply,
'HC705 programmers board, soldering
pencil, solder, and back issues of Micro
C.

The simulator for the MC68HC705C8
microcontroller that Karl writes so much
about will be the first of many articles
about microcontrollers and embedded
systems.

♦ ♦ ♦

MICRO CORNUCOPIA, #53, May, 1990 81

Commenting Disassembler!

SOURCES 486

■ SEE HOW PROGRAMS WORK

■ EASILY MODIFY PROGRAMS

SOURCER'" creates detailed commented
source code and listings from memory and
executable files directly suitable for reassem-
bly. Built in data analyzer and simulator re-
solves data across multiple segments and
provides detailed comments on interrupts and
subfunctions, I/O ports and much more. In-
cludes a definition file facility to include your
own remarks and descriptive labels, force
data types, and more. Complete support for
8088 through 80486, 8087 to 80387, and V20/
V30 instruction sets.

We welcome comparisons with any other prod-
uct, because no product comes close to the
ease of use and output clarity of SOURCER.
"Sourcer is the best disassembler we've ever
seen." PC Magazine 1/17/89 page 101 .

BIOS SOURCE

for PS/2, AT, XT, PC and Clones

■ CHANGE AND ADD FEATURES

■ CLARIFY INTERFACES

The BIOS Pre-Processor'" with SOURCER
provides the first means to obtain accurate
legal source listings for any BIOS! Identifies
entry points with full explanations. Resolves
PS/2's multiple jumps for improved clarity.
Provides highly descriptive labels such as
"video.mode" and much more. Fullyautomatic.

ASM ProPak,

■ ASSEMBLY TOOLS PACKAGE

Save $50 with the ASM ProPak™ assembly
tools package. It includes Sourcer with the
BIOS Pre-Processor; Unpacker,'" for unpack-
ing packed EXE files; ASMtool,'" to automati-
cally generate flowcharts and tree diagrams
from assembly source code; and View-It,'" to
quickly view files of any size and see up to 8
times more information on a single screen
using EGA/VGA equipment.

Sourcer-Disassembler $ 99.95

BIOS Pre-Processor 49.95

Unpacker-Unpacks packed files 39.95

ASMtool-Source code analyzer 89.95

View-lt-Zoomable file viewer 69.95

ASM ProPak-AII Above (save $50) 299.75

Sourcer w/BIOS-(save $10) 139.95

Shipping & Handling: USA $5; Canada/Mexico $10; Other Countries
$15; CA Residents add sales tax. Visa/MasterCard/COD accepted.

NOT COPY PROTECTED
30-DAY MONEY-BACK GUARANTEE

// within 30 days of purchase you find our product does
not perform in accordance with our claims, call our
customer service department and we will gladly
arrange a refund.

For orders or information, call:

1-800-662-8266

V COMMUNICATIONS, INC.

4320 Stevens Creek Blvd., Suite 275, Dept. MC
San lose, CA 95 1 29 (408)296-4224

Reader Service Number 62

Figure 1 — Kruskel's Algorithm Implementation

PROGRAM Kruskal;

{* Author : Michael Bunt

This source code is released

into the public domain. *)

type

keystr » STRING [51;

edgeptr «* A edge;

edge s» record

bptr, fptr : edgeptr;

keyl, key2 : keystr;

cost : REAL

end;

compptr s» A comp;

comp = record

bptr, fptr : compptr;

compname : INTEGER;

edge : edgeptr

end;

VAR

I, J, K, P,compcount ; INTEGER;

maincomp : compptr;

mainedge : edgeptr;

fill : text;

PROCEDURE readfile;

(* proc to read in data file and *)

VAR

(* create initial list *)

e : edgeptr;

BEGIN

assign (fill, 'KRUSKAL .OAT' ) ;

(* open data file *)

reset (fill) ;

(* set file pointer to beginning *)

if not eof (fill) then begin

(* if not empty file *)

new(e) ;

{* create new pointer *)

mainedge :— e;

(* set it to first edge *)

readln (f ill, e* .keyl, e A .key2,

e A .cost); (* read in edge data *)

e A .bptr •- mainedge;

(* set bptr to list start *)

e A .fptr :- NIL

(* set forward pointer to NIL *)

end;

while not eof (fill) do begin

(* while still not end of file *)

new(e) ;

{* create new pointer .*)

readln (fill, e A .keyl, e A .key2,

e A .cost); (* read in edge data *)

mainedge A . bptr := e;

(* insert at front of list *)

e A .fptr := mainedge;

mainedge := e

end;

close (fill)

(* close data file *)

END;

PROCEDURE writelst (p : edgeptr) ;

(* writes out any edge list *)

VAR

next : edgeptr;

cost : REAL;

BEGIN

cost : » . ;

(* set list cost to zero *)

next := p;

(* set work pointer to list start *)

while (next <> NIL) do begin

(* while not end of list *)

writeln (next*. keyl, next". cost: 5: 3, ' ' ,next A .key2) ; (*list data*)

cost :- cost + next A ,cost;

(* increment cost by current edge *)

next :- next A . fptr

(* inc edge pointer to next in list *)

end;

writeln ( ' Cost is ',cost);

(* output list cost *)

END;

PROCEDURE writecomp;

(* proc to write out all components *)

VAR

next ; compptr;

BEGIN

next :- maincomp;

(* start at first component *)

WHILE (next <> NIL) do BEGIN

(* while still components in list *)

IP (next A . edge <> NIL) THEN

BEGIN (* if component no empty *)

writeln {' Component ' ,next"

.compname); (* print component name *)

writelst (next A .edge)

(* output edge list for component *)

END;

next := next A . fptr

(* inc component to next in list *)

END

END;

PROCEDURE sortlst;

(* proc to sort initial edge list *)

82 MICRO CORNUCOPIA, #53, May, 1990

.(*

VAR

tcoat : REAL;
tkeyl,tkey2 : keystr ;
I, J, count : INTEGER;
e, cur, prev : edgeptr;

BEGIN

e :- mainedge;
count := 0;

WHILE e A .fptr <> NIL DO BEGIN
count := count + 1;
e :» e A .fptr
END;

FOR I := 1 TO count -1 DO BEGIN
cur :=> e;
prev := e^.bptr;
FOR J := count DOWNTO I DO BEGIN

IF cur A . cost < prev A . co at THEN BEGIN
tcost := cur A . cost;
cur A . cost := prev A .cost;
prev A .cost :- tcost;
tkeyl := cur A . key 1;
cur A .keyl := prev A .keyl;
prev". key 1 := tkeyl;
tkey2 := cur A .key2;
cur A .key2 := prev A .key2;
prev A .key2 : = tkey2
END;
cur :— prev;
prev := prev A .bptr
END
END
END;

(* start at beginning of list *)

(* then implement bubble sort *)

(* bubble sort is 0(n**2) so *)

this implementation of Kruskal's *)

(* would have to have a O(nlogn) *)

(* sort to be O(nlogn) *)

(* swap code *)

PROCEDURE buildlst;
VAR

cur : edgeptr;

compl, comp2 : compptr;

(* proc to build minimal spanning *)
(* tree if it exist *)

PROCEDURE find (key : keystr; (* proc to find the component that

compptr) ;

(* a key of an edge is in

VAR comp
VAR

found : BOOLEAN;
edge : edgeptr;
BEGIN

found : = FALSE; (* set found flag to false

comp := maincomp; (* start a first component

WHILE (comp <> NIL) AND (NOT found) DO (* while not empty
BEGIN (* component & key not found

edge := comp A .edge; (* start at first edge *)

WHILE (edge <> NIL) AND (NOT found) DO (* while not empty
BEGIN (* edge & key not found

IF (edge A . key l=key) OR (edge A . key2=key) THEN (*if key found*)
found := TRUE (* set flag *)

ELSE edge := edge A .fptr (* else increment to next edge *)
END (* while *) ;

IF NOT found THEN comp:=comp A .fptr (* if not found-next compon *)
END (* while *) ;

IF NOT found THEN comp:=NIL (* if still not found- no component *)
END;

PROCEDURE joincomp(VAR el, c2 : compptr); (* proc to join compons *)

VAR

1 : edgeptr;

BEGIN

1 := cl A .edge;

WHILE l A .fptr <> NIL DO BEGIN

1 :- l A .fptr
END;

l A .fptr := c2 A .edge; (*last in 1st list points to 1st in 2nd list*)
c2 A .edge A .bptr :- 1; (*lst in 2nd list points back to last in 1st*)
c2 A .edge := cl A .edge; {* 2nd comp points to 1st in 1st list *)
c2 A .edge A .bptr :— NIL; (* clean up back pointer *)

cl A .edge := NIL (* 1st list now empty *)

END;

(* set temp to first edge list

(* while not last edge in list

(* inc to next edge in list

PROCEDURE newcomp;
VAR

cur : compptr;
BEGIN

new (cur) ;

(* proc to create a new component *)

(* create new pointer *)

et organized
instantly!

Create, Find, View, & Edit
notes, addresses,
documents, data.

* No setup.

* No keywords to define.

* Configurable,

Full-screen editor.

* Turbo fast & little

IZARD

The Intelligent Information
Processor

Only $49

ACQUIRED INTELLIGENCE

PO BOX 2091 DAVIS, CA 95617

916-753-0360

Reader Service Number 72
MICRO CORNUCOPIA, #53, May, 1990 83

YOU WANT REVOLUTION
NOT DISK ROTATION!

Blue Flame II Solid State Disk

■ For the PC, XT, AT, '386 DOS computers, no mechanical wear

■ Revolutionary way to solve your disk access bottleneck -
Rebel against hard drive delays

■ Like a hard disk, but it's 10 to 20 times faster

■ Expandable - 1 to 8 MB per card, up to 32 MB per drive

■ EMS for Le$$ - LIM/EMS 3.2 & 4.0 emulation software available

■ Battery Backup option, makes the Blue Flame II non-volatile, $1 30

■ Concurrent DOS compatible too

■ 2-year warranty, parts and labor

1 MByte - $495 SemiDisk Systems, Inc.

2 MByte -$695 1 1 080 S.W. Allen #400
4 MByte - $call Beaverton. Oregon 97005
8MByte-$call f503| 626-3I04 FAX (503) 643-0625

Reader Service Number 162

Micro Cornucopia

Phone Orders:
<5Q3) 382-5060, M*F t 9 AM* PM PST

P£>< Box 22$, Bene*, Oregon, 97709

PC TIMER TOOLS

PCHRT is the definitive answer to execution pro-
filing and embedded timer requirements in a
PC/MSDOS environment. 46 functions manage
three distinct classes of high resolution timer
functionality:

21 functions implement multiple microsecond
timers. Timers may be stopped, started, sus-
pended, resumed, named, queried, and
complete reports of all timer activity may be
generated with a single function call.

16 functions manage microsecond timers that
may be inserted in virtually any PC interrupt vec-
tor. Interrupt timers may have results listed by in-
terrupt function requested, so BIOS, DOS, and
EMS interrupts may be profiled effortlessly!

9 functions implement precision microsecond
delays, perfect for interfacing data acquisition
and process control hardware where precise
timing is required.

PCHRT includes libraries for TC, TP, and MSC,
example programs, manual, full source, and is
$49.95 ppd USA, elsewhere add $4.00. VISA/
MC accepted. A 30 day "No Questions Asked"
money back guarantee insures your satisfaction.
Call or write for our seven page brochure.

RYLE DESIGN

POB 22, Mt. Pleasant, Ml 48804
517-773-0587

Reader Service Number 171

incrememt component name

(* set component name

(* clean up back pointer

insert new component a front of list

P :~ P +1;

cur A . compname := P;

cur A .bptr := NIL;

cur A ,fptr := maincomp; (*

cur A . edge := NIL;

cur A .fptr A .bptr := cur; maincomp := cur

PROCEDURE addedge (VAR edge : edgeptr;VAR comp : compptr) ;

BEGIN (* proc to move edge form intial list to component

mainedge : = edge A .fptr; (* set begin edge to next edge

edge A .fptr := comp A .edge; (* begin insertion

edge A .fptr A .bptr := edge;

comp A .edge := edge;

edge A .bptr := NIL (* clean up back pointer

END;

BEGIN

cur := mainedge ;

WHILE cur <> NIL DO BEGIN

f ind{cur A .keyl, compl) ;

find<cur A ,key2, comp2) ;

(* begin of BuildLst main *)

(* set work pointer to begin edge list *}

(* while still edges to process *)

(* find component that key is in *)

(* find component that key is in *}

IF (compl = NIL) AND (comp2 = NIL) THEN BEGIN (*if neither key*)
newcomp; (* found in components, create new component *)

addedge (cur, maincomp) (* add edge to new component *)

END;

IF (compl = NIL) AND (comp2 <> NIL) THEN BEGIN (*if key only in*)
addedge (cur, comp2) (* component 2, add edge to component 2 *)

END;

IF (compl <> NIL) AND (comp2 = NIL) THEN BEGIN (*if key only in*)
addedge (cur, compl) (* component 1, add edge to component 1 *)

END;

IF (compl <> NIL) AND (comp2 <> NIL) THEN

BEGIN (* if both keys found in components *}

IF (compl <> comp2) THEN

BEGIN (1

addedge (cur, compl) ;
joincomp (compl , comp2)
END
ELSE

mainedge := cur A .fptr
END;

cur := mainedge
END (* endwhile *) ;
END (* buildlst *) ;

if keys not found in same component *)

(* add edge to one of the components *)

{* join the two components *)

(* cycle found *)
(* discard edge *)

(* increment to next edge *)

BEGIN
P := 0;

maincomp :- NIL;
readfile;

writeln (' Input list ...');
writelst (mainedge) ;
sortlst;

writeln; writeln (' Sorted list. . . ' ) ,
writelst (mainedge) ;

buildlst; (* build the minimal cost spanning tree

writeln; writeln (' Minimal cost spanning tree..');
writecomp (* output the components

Data File: kruskal.dat

(* initialize component name *)

(* initialize component list *)

read in edge list from data file *)

(* output initial edge list *)

(* sort the initial edge list *)

(* output sorted edge list *)

)

2.0

8.0

10.0

14.0

8.0

15.0

17.0

12.0

4.0

8.0

14.0

4.0

13.0

12.0

10.0

4.0

16.0

84 MICRO CORNUCOPIA, #53, May, 1990

TIDBITS

By Gary Entsminger

P.O.Box 2091

Davis, California 95617

Finding Text Faster, An
Optimizing TSR

Development Tool, & TAP CIS

Searching for the ultimate search
routine? Gary looks closely at Power Search
and at searches in general.

Just when I thought I'd figured out
how to search a string for a pattern
(or substring) fast (see Tidbits, Micro
C, #52), Blaise Computing comes along
and blows my doors off. Their system,
called Power Search, is a search compiler
that finds text faster by building object
code at runtime for each search. A better
idea!

The object code, executed as a func-
tion, gets the address of a buffer to scan,
the buffer length, and the pattern, then
goes to work. When I added the com-
piler to Wizard, my search model from
last issue, its speed improved consider-
ably.

Depending on the length of a search,
the search compiler improved Wizard's
performance by over 20% from its sec-
ond best effort (using a good lookup
table) and by 120% (using unoptimized
brute force).

For those of you who missed last
issue, the brute force search does noth-
ing fancier than begin at the beginning of
a string and move through character by
character looking for a match. Lookup
table algorithms look for ways to elimi-
nate some of the target string from the
search.

Refer to Figures 1 and 2 for a com-
parison of four search methods:

(1) brute force;

(2) optimized brute force;

(3) table lookup system (Boyer-
Moore);

(4) Blaise's search compiler.

In Figure 1, we're searching through 1

meg. In Figure 2, we're searching
through 3 megs. For more details about
methods 1, 2, and 3, see last issue. In the
next few paragraphs I'll try to show you
why a search compiler is faster.

A typical search through a buffer for
the occurrence of a string might take the
following approach (in C):

#define LEN 327 67 /*32K buffer*/
#define PATLEN 20

char Buf [LEN] ;
char Pat [PATLEN] ;
unsigned i, j;

for (i = 0; i < LEN - PATLEN; i++)
{

for (j = 0; j < PATLEN; j++)
{

if (Buf[i+j] != Pat[j])
break; /* no match! */

if (j = PATLEN)

break; /* match at (i) */
}

Note that most of the code (and time)
is used to manage the array indices.
Power Search gains speed by eliminating
this overhead, by (effectively) hand
coding each lexical element of a pattern.
This hand coding builds a production (or
set of bytes that correspond to machine
instructions) to scan the buffer for the
pattern.

Each time a user submits a pattern for
a search, the search compiler builds this
object code — at runtime very, very fast.

To get an idea of what Power Search
is up to, consider the following example
(from the Power Search manual). In this
example, a user submits a character that
she wants matched (either upper or
lower case). First, a typical method to

Figure 1 — Effects of the Search Method on Speed (1MB of Files)
25

w
■o

c
o
o

CD
CO

brute

opt. brute B-M

Search method

compiler

MICRO CORNUCOPIA, #53, May, 1990 85

compare characters, hand-optimized in
assembly.

The method uses the following as-
sumptions:

(1) DS:SI points to the next character
to scan;

(2) ES:DI points to the next pattern
character;

(3) The direction flag is cleared;

(4) AL can be scratched.

We can't use EPE CMPSB because we
want the comparison to be case-insensi-
tive. So the code looks like this:

lodsb ; (AL) = buffer char
xor al,es: [di] ; (ES:DI)=pattern char
inc di /advance pattern char

;do they match?
je match ;yes; exactly,
cmp al, 20h ;case problem?
jne failure ;no; they just don't

; match .
match:
; next production .

Now, the search compiler's op
timized code:

lodsb

cmp al, 'X'
je continue
cmp al, ' x'

; (AL)=next buffer ch;
;upcase match?
;yes. next production
;lowcase match?

jne failure ; no .

continue :

;next production.

"Both methods have to load the next
character from the buffer into AL with a
LODSB, a fast one-byte instruction. But
the general method has to read the pat-
tern character to compare with the buff-
er Character, and to increment the
pointer to the pattern (DI). These two
steps are unnecessary in the search com-
piler's production because the character
is known when the search begins."

In effect, the Power Search method
leaves one compile until runtime. This
last compile is blazingly fast — fast
enough to increase search speed signifi-
cantly on long searches.

You can add any of the seven func-
tions to your C code very easily. Blaise
gets you started with several examples
and precompiled libraries for Microsoft
and Turbo C. The manual includes a
function reference and an illuminating
discussion about how the search com-
piler works.

Although I'm impressed (can you tell
yet?) with the speedy string search

speed I've discussed, I'm selling this
program short. Power Search can also
handle regular expressions, including
negated character classes, subexpres-
sions, and repetitions. A regular expres-
sion, for you noncompiler-writing types,
describes a language. I won't go any
farther into that here. For more informa-
tion, check one of the classics, such as
Principles of Compiler Design by Aho and
Ullman.

In short, you can easily incorporate
functions to handle regular expressions
into your C applications.

If I'm leaving you with too positive
an impression about Power Search, I'll
throw in one complaint — one I'll direct
at many other companies as well as
Blaise. The manual's index isn't bad, but
it's incomplete. An index should be use-
ful most of the time. If not, users won't
use it! 'Nuff said.

The CodeRunner function library in-
cludes:

• tools for installing and (safely) unin-
stallingTSRs;

• support for up to 256 TSR entry
points (so you can allow users to go
directly to specific parts of an appli-
cation);

• an event scheduler (for timing
popups);

• and an ultra-compact BCD math li-
brary.

TSRs developed with CodeRunner
can be very small (<2K) because
CodeRunner lets you discard initializa-
tion code when it's no longer needed.
You put your startup display, messages,
etc., in a disposable data module and the
code and data you want to remain resi-
dent in a permanent module. Then using
a 4 function sequence similar to the fol-
lowing, you set the TSR in motion.

Figure 2 — Effects of Search Method on Speed (3MB of Files)

brute opt. brute B-M

Search method

compiler

For more information:

Power Search $149
Blaise Computing
2560 Ninth St., Suite 316
Berkeley, CA 94710
(800) 333-8087

Optimized TSRs

You C'ers searching for a concise
package of TSR tools should check out
CodeRunner from Microsystems Soft-
ware. CodeRunner is a compact library
of over 200 functions written in as-
sembly language that you can call from
your Turbo C and Microsoft C pro-
grams.

/♦Define stack, enable TSR on ret*/
stay_resident (isr_stk, STK_SZ*2) ;
/* Enable init data disposal */
idata_end=£init_data_end;
/* Enable init code disposal */
icode_beg=init_code_start ;
/* Install hotkeys */
install_hk (hk_list, service,
2*STK_SZ,3) ;

The CodeRunner package includes a
template for creating TSRs, tips on op-
timizing C programs, and several useful
examples which include source code.
The Microsystems Software BBS (free
when you register) also has additional
source code and tips from CodeRunner

86 MICRO CORNUCOPIA, #53, May, 1990

developers. The CodeRunner manual
(here I go again) unfortunately is a little
too concise. It's obviously intended for
seasoned C programmers. If you're not
one, expect to work a bit getting a TSR
up and running.

If you're looking for a professional
development tool that'll let you create
compact, fast TSRs, CodeRunner might
be your ticket.

It's $149 from:

Microsystems Software
600 Worcester Rd., Suite B2
Framingham, MA 01701
(508) 626-8511

Note: Microsystems Software specializes
in developing software for the disabled. If
you need information of this sort, get in
touch with them.

TAPCIS

It took Bruce Eckel (and others) a long
time to convince me to go ONLINE
nationally. I had my reasons. In particu-
lar, I hate rummaging around bulletin
boards; I'm neither a shopper nor a
swapper. So seven years after my first

bout with a Kaypro, I joined Compu-
Serve. It was more or less what I ex-
pected: forums, marts, conferences,
weather reports, etc.

Bruce had argued that EasyPlex alone
(CompuServe's electronic mail service)
was worth the price of admission. My
first few experiences ONLINE suggested
otherwise. I figured out how to send and
receive mail, but it was more trouble
than it was worth, and certainly not
worth the price of admission. Then I dis-
covered TAPCIS.

TAPCIS is an access program for
CompuServe that does indeed take the
pain (and time) out of sending and re-
ceiving mail ONLINE. It's fast and vir-
tually automatic.

I run the program, enter N (for New),
and TAPCIS calls CompuServe, logs on,
checks mail, downloads any that's wait-
ing, and logs off while I make coffee.

To upload a file or a message, I tell
TAPCIS where the message is on my
hard disk, and it logs on, uploads, logs
off, etc., while I drink my coffee.

TAPCIS, you're wonderful. If you use
CompuServe, you can download it and
play free for 21 days. If you decide to

keep using it, you register your copy for
$79 and get a manual and the latest ver-
sion. You can even have TAPCIS register
for you ONLINE. Now that's friendly.

For more information:
Support Group, Inc.
Lake Technology Park
P.O. Box 130
McHenry, MD 21541
CompuServe 74020,10 (Go TAPCIS)
(800) USA-GROUP

And, you can reach me on Compu-
Serve, 71141,3006

Postscript

As the sun sets, Tonto and the
Masked Man ride off on their beautiful
horses to cook a beans and bread supper
in the old west. With a "Hi ho Silver,
away," I'm reminded of another great
saying — Thanks for all the memories.

It's been good getting to know you
Micro C'ers. Let's meet again someday —
at a soggy get-together, a conference, or
on some other pages. And that, friends,
ends Tidbits.

♦ ♦ ♦

Turbo Pascal/Turbo C Users:

Don't Miss
The Boat!

Just knowing where to start can be a challenge. And once
you have experience -where do you look for inspiration?

Since 1984, TUG has offered its members access to a
worldwide network of thousands of Turbo Pascal and
Turbo C users through TUG Lines, its lively, 48-page
bimonthly journal. Tutorials and help for beginners. Tips
and techniques for experienced programmers. Bugreports
with workarounds. Product reviews. Volunteer
consultants you can call for answers.

But TUG is much more than a newsletterl Membership
options include a multiuser BBS, videos, contests, an
extensive library of hand-picked and tested software,
and an annual "GeTUGether" where experts gather to
share their knowledge at all levels.

Turbo User Group

Post Office Box 1510 • Poulsbo, Washington 98370
(206) 779-9508

Annual dues (US funds): U.S. $27.00; Canada $32.00; Overseas $39.00

FREE ISSUE!

Micro C readers are invited to receive a
FREE sample issue of Midnight Engineering.

The Journal Of Personal Product
Development provides a unique mix
of practical hardware and software
engineering mixed with the business

aspects of small scale product

development. Your guide to developing

and marketing your own products.

Midnight

Engineering

Suite 7041

111 E. Drake Rd.

Ft. Collins, CO 80525

303-491-9092 (anytime)

MICRO CORNUCOPIA, #53, May, 1990 87

SHARE
WARE

My Favorite Tools

Helpful Shareware

By Nelson Ford

The Public (software) Library

P.O. Box 35705

Houston, TX 77235-5705

(713)665-7017

CIS: 71355, 470

FAX: (713) 524-6398

In this, his first shareware Micro C
column, Nelson covers his favorite utilities.
Great stuff.

Prior Shareware columnist Anthony
Barcellos is going to be a tough act
to follow. For the last two and a
half years, Tony has done an admirable
job of coming up with the real gems.

Over the past eight years, I've tested
and reviewed thousands of public
domain and shareware programs for the
Public (software) Library.

I've written a dozen shareware pro-
grams and am one of the founding mem-
bers of the Association of Shareware
Professionals.

Terms

First, let me clarify a few terms.

When I say "pd/shareware," I in-
clude "freeware," too. Public Domain is
software which the author has donated
to the public domain. (The absence of a
copyright notice is not sufficient to make
a program pd.)

Freeware used to be the same as
shareware. Andrew Fluegelman trade-
marked the name freeware, so other pro-
grammers invented the name shareware.
After Fluegelman's death, the term
freeware fell into generic use, meaning
software that is free, but copyrighted, as
opposed to free and publicly owned.

Since most of you spend your time on
DOS machines, I'll start out with a dis-
cussion of. . .

The Ten Greatest DOS Tools In The
History Of Man

I would hate to think of having to
work my 14+ hours a day on a computer
without the following:

(1) LIST, by Vernon Buerg. You prob-
ably already know about this file view-
ing utility, but you may not know that
Buerg keeps making it better. If you
haven't seen version 7, it's time to dial
up your favorite BBS.

(2) FGREP, by Christopher J. Dun-
ford, is the world's fastest text search.
Sure, you can get a fancy TSR like Gofer,
but you can't beat FGREP for speed and
convenience.

(3) SD-whatever, anyone's sorted
directory utility is better than what DOS
provides. I prefer HotDIR, or one of the
others that color-codes file names by ex-
tension. It's easy to spot all the OBJ files
in a crowded directory.

(4) MOVE FILES 4.2, by Bryan Hig-
gins, helps you move files between sub-
directories or disks. After years of
struggling with deficient file movers, it's
great to find one that's very close to per-
fect. The only question is why it took so
long to show up.

(5) CED, by Christopher Dunford, is
to the DOS command line what a full-
screen editor is to Edlin. It lets you "cur-
sor up" previous DOS command lines
and edit them using all the editing keys
(e.g.: home, end, word left/right). CED
is freeware, but there's a commercial
version that's even better.

(6) HindSight, by Christopher Dun-
ford. Anyone who works in DOS needs
to see what's just scrolled off the screen.
Though there are several video back-
scroll utilities, HS works with Dunford' s
CED, so I use it. Also check out FANSI-
Console, which includes a wealth of
video related features along with back-
scroll.

(7) LHarc, a freeware file archiver
from Japan. LHarc is the best file com-
pression I've seen. Some files will com-
press to a fifth their original size.

(7a) I must sneak a related cleanup
utility in here. LHdel, by Duane Hen-
dricks, examines the contents of an
LHarc archive file and deletes matching
files in the current directory.

(8) Newkey, by Frank Bell. If you
don't have a keyboard macro program,
you need Newkey.

(9) DESQview. Oops; okay, this mul-
titasking program isn't shareware.
Worse yet, I lied in (8): I don't use New-
key anymore, since DV can automat-
ically load a different set of macros each
time you open a window. Since I only
use DV for task swapping, you might
think that Software Carousel could do
just as well. But DV has too many nice
extras to give up.

(10) SitBack, by SitBack Technologies.
This TSR auto-backup utility isn't
shareware either, but I am the world's
worst at backing up my work and
there's no shareware solution. SitBack
will backup preselected files in the back-
ground at specified times. This TSR uses
a modest 16K.

The closest shareware equivalent is
AutoSave, by Biologic Corp. It records
your keystrokes to disk, plus it saves
your work for you as often as you wish.

How To Write Shareware And Make A
Million Dollars

Step One: Write a shareware pro-
gram. Step Two: Win a lottery.

A half-dozen shareware authors have
been able to combine the two. Tom
Smith and his partners turned a
shareware communications program,
Procomm, into $15 million a year. Jim
Button (PC-File) and Bob Wallace (PC-
Write) make over $2 million a year.
Marshall Magee (Automenu) and Dave
Berdan (File Express) are grossing over
$1 million a year.

88 MICRO CORNUCOPIA, #53, May, 1990

These big bucks have put stars (and
dollar signs) in the eyes of thousands of
programmers. That's why thousands of
people make little from shareware while
a handful prosper. If you have ideas
about making real money from
shareware, consider these:

(1) Writing the program is the easy
part. Sending out disks to vendors,
uploading to BBSs, responding to users,
fixing bugs, adding requested features,
and improving the documentation are
not only more work, they aren't as much
fun. They take up time and cost you
money.

Two classic examples of proper
marketing are Datastorm's Procomm
and Magee's Automenu. Procomm was
not the first shareware communications
program, nor was it significantly better
than Qmodem. But Datastorm's people
knew how to market the program.

Marshall Magee took a program in
one of the most crowded arenas of PC
software — DOS menu programs — and
turned it into a winner. In shareware
circles, Magee is known as Mr. Image.
He knows how to market, and he
learned it on his own.

(2) You don't have to choose between
shareware and nonshareware market-
ing. There is nothing sacred about
shareware; it's just another way of
marketing. As the prices of retail pro-
grams have approached shareware
prices, programmers have had to be
aware of both arenas.

If you market through both retail and
shareware channels, don't cripple the
shareware version. It's better to sell re-
tail only than to pawn off a crippled
shareware program.

(3) Write an outstanding program
that businesses can use. Businesses are
still the top shareware supporters.
Games and DOS utilities may bring in a
few bucks, but that's about all. Yet
Another Text Editor or Yet Another Flat
File Database Manager is not likely to
generate much cash, either.

(4) Don't reinvent the wheel. If you
want to rewrite the same old utilities
over, fine. If you want to make money,
know what's already there and be pre-
pared to either out-program it, out-
market it, or forget it.

(5) Hang in there. Many people write
a program, get all excited while they
send the program out and upload it to
BBSs, and then get discouraged and quit
when a month or two goes by without
hearing anything. The shareware mill

grinds exceedingly slowly.

Eventually, your program will be-
come ingrained in The System of
pd/ shareware distribution. If you have
a good product and you continue to im-
prove it and promote it, you can profit
from it for years. I still get weekly regis-
trations from a 1984 version of Diskcat at
a P.O. Box that I have not publicized in
over three years.

None of the top moneymaking
shareware authors of today made much
money their first six months. It takes that
long for programs to start getting passed
around. If you need quick income,
advertise. Unfortunately, advertising's
not likely to be worthwhile.

(6) Shareware is not a way to make
money from inferior software. Some
folks try marketing their program as
shareware because it's not slick enough
to sell on the commercial market. They
shouldn't bother.

In theory, shareware has to be better
than retail-only software because people
get to try shareware before paying for it.

I could go on for pages and pages. In
fact, I already have in the Programmers
Guide, an on-disk guide for shareware
programmers. It contains tips on how to
polish a program and documentation,
how to protect your software legally,
how to find necessary supplies and serv-
ices at the best prices, how almost anyone
can get a MC/VISA merchant account,
and more. The Programmers Guide is on
CompuServe (GO IBMPRO) as
GUIDE1.ARC and GUIDE2.ARC. You
can also get it by sending $5 to my
address and asking for the Programmers
Guide disk.

The Tax Man Cometh

By the time you read this issue, you
should be starting to think about your
tax return. My longtime favorite tax pro-
gram, commercial or shareware, is AM-
Tax.

PC-Tax is another veteran of the
shareware world. It usually supports
more forms than AM-Tax, but it's line-
oriented, an approach which is more ap-
propriate to a teletype terminal than a
full-screen computer.

AM-Tax uses a full-screen, fill-in-the-
form approach in which you scroll
through replicas of the tax forms, fill in
the blanks, and watch AM-Tax do calcu-
lations and carry totals forward for you.

AM-Tax works so intuitively you'll
probably never read the manual. Any
time you need a subschedule, all you

have to do is press F6. If no IRS sub-
schedule exists, it creates a worksheet.
After you use the worksheet or sub-
schedule, you can return to the original
schedule bringing along the balance.

In contrast, the last time I tried Turbo
Tax, a popular retail-only tax program, I
had to work through a confusing form
menu to create a subschedule. Arbitrary
worksheets were not an option. I got so
bogged down in Turbo Tax that I never
could complete the tryout.

The shareware version of AM-Tax
supports form 1040 and schedules A, B,
C, D, E, F, R, and SE. All printouts can
be filed directly to the IRS. If you need
more schedules, corporate or state
forms, etc., they have more advanced
versions.

Finding These Programs

All pd/ shareware programs men-
tioned in this column are available from:
public BBSs, CompuServe, user groups,
and disk distributors such as PsL. In par-
ticular, PsL coordinates quite a bit with
IBMNET on CompuServe.

IBMNET is divided into many differ-
ent Forums and each Forum has many
different Data Libraries. The quickest
way to find a program on IBMNET is
through the File Finder (GO IBMFF).
Once there, you can search by program
name or Key Word. For example, to find
AM-Tax, you could search for the name
or the Key Word "TAX."

CompuServe limits file names to six
characters plus a three-character exten-
sion, so longer names may be abbre-
viated in ways you could never imagine.
When this happens, the actual program
name will often be recorded as one of
the Key Words, so search for the pro-
gram name in the Key Word field.

References

Chip Merchant

9285 Chesapeake Dr., Ste. L9285

San Diego, CA 92123

(800) 426-6375

(619) 268-4774

SitBack Technologies
9290 Bond, Ste. 210
Overland Park, KS 66214
(913) 894-0808

♦ ♦ ♦

MICRO CORNUCOPIA, #53, May, 1990 89

8088 Hardware Development

TECHTIPS

Aiding The Ailing 8088

I discovered a way to diagnose a recalcitrant
8088 recently while working on controller cards.
(It turned out that the steps outlined by Ed
Ninsley for the 8031 in a back issue of Circuit
Cellar INK can be modified for use with 8088s.)

My situation was pretty down and dirty:
building an 8088 board and trying to get it
working. I couldn't even get the 8088 to execute
a single instruction.

Ed's steps boil down to the following.

• Disconnect the data bus from all memory
devices and ground all the data lines. This
forces the instruction ADD AL,[BX] onto
the bus.

• Ground both of the interrupt pins (NMI
and INT) and also the HOLD request pin
(used by DMA and things like that).

This way the CPU will free-run, in-
crementing addresses through segment FFFF.
Remember that most of these accesses will wrap
around to 00000; only the first 16 bytes come
from the high end of memory.

Basically this gives you the ability to make
sure that address latches and CPU clock are
working before you add more things (such as
memory decoding and memory).

Nathan Engle

6465 Piping Rock Lane

Indianapolis, IN 46254

High Memory Update

Since the original incarnation of my high
memory card ("Filling In The Holes On Your
XT," Micro C, Issue #52), I've rearranged and
added things to my computer. Specifically, I
added an expanded memory board.

Expanded memory is one of the greatest
things since sliced bread, but it does require 64K
of open memory space. For me, the simplest
place to map it was in block D (D000:0000 to
D000:FFFF). With the commercial memory cards
referenced in the article, this would mean sepa-
rate cards if you wanted to fill in the remaining
A and E blocks.

The SRAM card, however, may be modified
as follows:

• Interchange address line A18 and A17 at
the I/O bus interface;

• Move the chip select lines from pins 12 and
13 of the '138 to pins 14 and 15. (See Figure
1.)

This mod will give 128K, split between the A
and E blocks. You might also want to rethink
just which blocks, if any, you want to battery
back-up. The change to E-DISK to make it a 64K
RAMdisk is relatively trivial.

Also, in the eternal search for extra memory,
a special note to Hercules (and clone) card
owners. When operating in text mode, it is
simply not possible to make the video access
anything beyond the first 4K.

Let's call this an overly slavish emulation of
the IBM mono adapter. When running one of
the CGA simulation programs, only the top 32K
is accessed.

Notice the hole between 4K and 32K
(B000:1000 to B000:7FFF). This memory gets
used only when doing native Hercules graphics.
For ordinary text applications, this 28K sits
there idle. Scandalous! Quickly, load something
up there and use it.

While on the subject of the HILOADing soft-
ware mentioned in the article, I must say it's one
of the more useful utilities I've ever written. But
you should use some common sense in its appli-
cation.

Some TSRs are designed to speed up various
PC functions, such as video BIOS accelerators.
Since they would be executing out of I/O bus
memory, which is typically slower than mother-
board memory, it doesn't make sense to load
them up there, even though it's possible.

And, some TSRs (cache programs come to
mind) are smart enough to "look around"
where they are. From their point of view there
isn't much memory left, so it wouldn't make
much sense to install themselves. If they refuse
to load, it may very well be something like that.

Finally, in the never-ending saga of living
with SideKick, I've gleaned a few more tidbits

90 MICRO CORNUCOPIA, #53, May, 1990

Micro Ads

Experience Embedded
Power

with the new Control-R II 8031 ^Controller Module

V 5 volt single supply operation (even with RS232)

V Data/Address/Control Bus Headers for expansion

V Direct Access to B031 Ports 1 and 3

V 8K on board EPROM space

V Socket for 8K of on-board Static RAM (6264)

V MAX232 socket to provide RS232 compatible port

V Memory decoded to allow use of 8052-BASIC jtP

V Assembled and Tested, not a kit

V Compact size: 3.5" x 4.5"

Control-R II, Enhanced 8031 Module
(SRAM & MAX232 optional)

Control-R II, Bare PC Board w/docs

64.95

25.00

Control-R I, 8031 Module 39.95

(Port 1 & 3 access and MAX232 option only)

Control-R I, Bare PC Board w/docs

8K Static RAM I.C.
MAX232 for serial I/O

14.00

10.00
6.95

PseudoSam 51, 8031 family cross-assembler 50.00
for MSDOS computers (5.25 Inch disk w/docs)

Shipping $3.00 US / $5.00 Canada
Illinois Residents Add 6.25% Sales Tax

Cottage Resources Corporation

Suite 3-672C, 1405 Stevenson Drive

Springfield, Illinois 62703

(217)529-7679

Reader Service Number 158

NEW!

Now Micro Cornucopia has microform
availability; 16mm microfilm, 35mm microfilm,
105mm microfiche copies of issues and articles
provided through...

University Microfilms International.

300 North Zeeb Road, Ann Arbor, Ml 48106-1346

313/761-4700

XSPAWN replaces the standard C spawn functions
with functions that can transparently swap most of
the parent process to expanded memory or disk
before executing the child process. The XSPAWN
functions are as easy to use as the standard C
spawn functions. Indudes all source code and
libraries for Microsoft C and Turbo C.

Price $79.95 plus $4 shiping and handling,
residents add sales tax.

Whitney Software, Inc.

P.O. Box 4999

Walnut Creek, CA 94596

415-933-9019

Reader Service Number 164

Learn a Modern Language

Our tutorials use a unique method to teach
you to program in a modern language. Pascal,
C, C++, Modula-2, beginning Ada, and
advanced Ada tutorials are available for
MS-DOS. Each tutorial is $39.95 + $3.00 P&H
(in US). M/C or Visa accepted.

Coronado Enterprises

12501 Coronado Ave NE, Albuquerque,
NM 87122

(505) 293-5464

Reader Service Number 206

Technology for the Arts

Vision- 16 Video Digitzer displays
32,768 colors, w software $1495
Willow VGA-TV w NTSC Output $495
Publishers' VGA/Frame Grabber $595
Panasonic CCTV B&W Camera $225
MIDI (music) Interface $119
MIDI Programmers Toolkit $39
Cakewalk 3.0 Sequencer $150

joel Sampson Engineering

P.O. Box 550363 (214) 328-2730

Dallas, TX 75355 BBS 328-6909

Reader Service Number 176

HEXED.

Get HEXED ^The HEXadecimal EDitor

Display and edit ALL characters in AW Y file.

• Full-featured, menu-driven binary file editing

• Simplicity and power ot a text editor

• Split-screen, simultaneous Hex value\ASCII symbol editing

• Easy access to file data formats, headers, etc.

• For all PCs & compatibles; 256K; DOS 2.0+

• $49.95 USA, Canada. $59.95 elsewhere.

Ideal for engineering, program development, hacking.

_ Call for FREE Demo Disk.

inc. 141 Peyton Rd. Sterling, Va. 22170
703*444. TREX

Reader Service Number 218

DSP32C COPROCESSOR BOARD

25 MFLOP Floating Point DSP

High speed NUMERICS and GRAPHICS

32 BIT on board data bus

16 bit PC/AT interface

ALL ON BOARD MEMORY DUAL PORTED

Parallel and codec serial IO

15 ms 1024 point FFT

Assembler, monitor, and libraries

SYMMETRIC RESEARCH

5 Central Way, Suite #9, Kirkland, WA 98033

(206) 828-6560

Reader Service Number 182

Searching for that IMSAI?
Need to sell your 286 to buy 386?

Private Listings is a nntireml publication for computer and
c omp u t e r related items for sale by private owners. Savings
galore on all types of recent and vintage items! Home of the
$5 two line one year adl

Call or write for a freo subscription!

Gateway Technologies, Inc.

180 Church Road

King of Prussia, PA 19406

(215) 354-0330

Reader Service Number 216

WISDOM OF THE AGES

First electronic book of quotes, sayings & ideas
brings over 1,000 of the world's greatest minds
to PC screens & printers.

Select a subject. Use the vivid parade of timeless
knowledge to act, write & speak better; earn
more.

Subjects are organized into sections to give you
the most benefit in the least time. Uses Filters to
customize output; Dynamic mode to stimulate
creativity.

It's never too early to get a head start.
Introductory Offer: $79.00 for all 5 disks.
30 day money-back guarantee.

Requires 256K, DOS 2.0 or later & 2 floppies or hard
disk. Add $2.00 S/H. CA residents add 7%. Outside
U.S. add $10.

MCR Agency, Inc.

6116 Merced Ave. #8 IMC
Oakland, CA 94611
1-800-767-6797
FAX 415-444-6561

Reader Service Number 181

8748 EMULATOR

Simulate your 8748 programs in software before
burning the EPROM. See all internal registers
and i/o pins. Script files can simulate complex
external events. Includes source $15

Other Products

DOS Source <mc/cic

Fullycommented1.1,2.1 olO/q>40

Hercules Tools e , c

Graphics stuff w/source $13

Math Library tf . c

ascii Floating pt. w/source *10

Information Modes

P.O. Drawer F

Denton, TX 76202

817-387-3339 1-80O-628-7992

iQoQeS

Reader Service Number 149

EGA FRACTAL MASTER

Whether fractal novice or fractal hacker, this easy to use
yet powerful package lets you explore:

The Mandlebrot set Julia sets

Dragonland Self-squared dragons

Biomorphs Newton's method

Peano curves Fractal landscapes

Interrupt, save and resume plots at any time. Easily
change colors, set boundaries, animate colors, create
mirror images. Supports fast integer computations (even
WITHOUT an 80386) and 80x87 math co-processors.
Includes a slideshdw program. Specify 5 1/4 or 3 1/2 disk.

Only $25 (postage included)

Paul W. Carlson

602 North Avenue, #23

Burlington, Vermont 05401

Reader Service Number 185
MICRO CORNUCOPIA, #53, May, 1990 91

A18

O 2

74LS00

lAIJJ

ADDROK

4 5

Original Decoding Circuit (Partial)

A18

ADDROK

IA17::::

Modified Decoding Circuit (Partial)
Figure 1 — Decoding for A and E Blocks.

of its inner workings. Yes, it does use
undocumented INT 21 h calls (function
34h, at least: the INDOS flag) but many
TSR writers do that. While surprising at
first, I now use it myself!

Second, it seems that on every clock
tick SideKick scans the interrupt vector
table. If it sees someone else grabbing an
interrupt it wants, it rearranges things
its way. Every clock tick! Talk about
chutzpah!

Oh yes, SideKick is unloadable. Don't
try it, though, if you've HILOADed it. A
guaranteed lockup.

Larry Shannon
5615 Truscott Terrace
Lakeview, NY 14085

Comment Comments

I agonized on whether to submit this
as a technical tip, as it is merely an un-
documented feature of a popular com-
piler. I decided to go ahead on the basis
that it is a potentially useful feature that
may increase programming productivity
(but I may be overstating its usefulness).

I've always found it somewhat awk-

ward to enter C-style comments with
their slash-star, star-slash pairs (/* */).
So I was quite surprised to find out that
Microsoft QuickC 2.00 accepts C++ style
comments! This feature, as far as I can
tell, is completely undocumented in the
QuickC manual. I haven't been able to
verify this feature in MS C 5.1. Have any
MS C owners out there seen this feature?

I stumbled upon the feature while
looking through the on-line help system.
It's only used in one place: the sample
program LOCK.C (to comment out two
lines of code). There are no other refer-
ences to C++ comments in the help sys-
tem. (Note: C++ comments consist of a
single slash-slash pair (//) with the
comment text to the right, similar to as-
sembly comments.) I've tested this fea-
ture with flawless results.

Things like this make you wonder if
the programmers at Microsoft don't
have a few other things they haven't told
us about. Enjoy!

Steven Byrnes
10510 Emnora
Houston, TX 77043

//Editor's comment: Larry discovered
C++ comments in FRACTINT (see the
Shareware column in Issue #52) when con-
verting the code to compile under Turbo C.
FRACTINT was written in MS C 5.1, so
there's your answer, Steven.

Friday The 13 th !

I spoke to you Micro C folk a few
weeks ago about some odd happenings
on a Haupage 386 mammy board: that
was, it appeared to kick into low gear
approximately 30 minutes after boot.
Well, I've tracked that dog down, and
had my suspicions confirmed by one of
the Houston area gurus.

Lo and behold, the poor thing had a
virus. The infamous Friday the 13*
virus, according to the guru. It's difficult
to chronicle the sequence of events as
nothing made any sense until it was
over. I ignored the first law of trouble-
shooting (well, all right, the third law; it
was (1) plugged in and (2) the power
supply was fine). I should have known
better when I saw that interrupts 08h
and 21h were hooked. Unless you're
running a legitimate program, there's no
excuse for them to be hooked.

Anyway, to make a long story short,
here are some of the clues for sniffing
out the beast:

1) Interrupts 08h and 21 h get hooked
from the get-go. No need to run any-
thing first. It gets into the boot code.

2) It expands uninfected COM and
EXE files as soon as they are run. Note
that it's sneaky about this as it restores
the original time and date of the file, so
no clues via time stamp.

3) In every case, the string "sUMs-
Dos" was embedded in the infected files.
Using the Norton Utility disk search or
grep are two ways to find victims.

There may be easier ways to rid the
beast, but I found DOS's FORMAT effec-
tive, saving the agony of real low level
formatting (I can't always remember
G=C800:5, or something like that). Any-
way, there ye have it. May the fleas of a
thousand camels invade the armpits of
the worms that invent these demons.
Disinfectedly yours,

Dave Stojan
10310 Lybert Rd.
Houston, TX 77041

Determining Required Stack Size

I used to find it difficult to determine
the required stack size for my C pro-
grams. Then it occurred to me that the

92 MICRO CORNUCOPIA, #53, May, 1990

Micro
Ads

PSFX TSR prints any Epson FX-85 or IBM
Graphics Printer compatible output as PostScript,
with graphics and IBM screen characters. $85.00 |
Also available for Novell's NetWare.

PSPIot Convert HP-GL to PostScript or EPS,
any line width or color. PostScript oriented editor
with test downloader and error display. ASCII file
print, any font or size. AutoCAD to PostScript.
Batch download of PC and MAC fonts. $175.00 |

EPScreen Capture PC screens as tiny EPS files,

frames optional. Includes IBM screen font. $99.00 I

Legend Communications, Inc.

54 Rosedale Avenue West

Brampton, ON, Canada L6X 1K1

(416)450-1010 30 day guarantee (800)668-7077

Reader Service Number 198

FIGHT PIRACY

&
PROTECT YOUR
^J PROGRAM $$$'s!

Since 1986, companies worldwide have been choosing Az-Tech
security products. II you demand the strongest protection available,
why not choose one of these "proven leaders":

• EVERLOCK Copy Protection • EVERKEY Hardware "Key"

• EVERTRAK Software Security Software Security

For IBM and Compatibles. 30 day money back guarantee.

Free info and demo disk available.

y*"*S_ Az-Tech Software, Inc.

\AZJ? 305 East Franklin

\/ Richmond. MO 64085

(800) 227-0644 FAX (816) 776-8398 (816) 776-2700

Reader Service Number 197

XenoCopy-PC $79.95 +S /h

PC-DOS program lets your PC

Read/Write/Format

over 350 formats

XENOFONT $49.95 +s/h

high quality text screen printouts

ideal for use in software documentation

Bold face and reverse video supported.

XenoSoft

2210 SIXTH STREET

BERKELEY. CA 94710

415-644-3366

Reader Service Number 39

\Please see us at Electro Booth #1121

IN CIRCUIT EMULATORS
8051 FAMILY

68HC11

CALL OR WRITE FOR FREE DEMO DISK AND
BROCHURE!

^^^__ B— ^_ - 51 East Campbell Avenue
flllHTll I (408)866-1820
■ ■**■ ■•■■W FAX (408) 378-7869
CORPORATION Campbell, CA 95008

Reader Service Number 186

Get

3 1/2 or 5 1/4 Disks
Available

All the
Software

in Issue 51
on Disk!

$6 ppd (U.S.) or
$8 ppd (Foreign)

1-800-888-8087

(orders only)
or (503) 382-8048

Hercules
Console Driver

Fully emulated text output in graphics mode on
your Hercules Graphics Card. Great for pro-
grammers debugging & working with graphics
programs. Write directly to the screen or run
DEBUG without leaving HGC graphics mode.

Just $24

(Shipping &c Handling Included)

Winter City Software

8723 - 162 St

Edmonton, AB Canada T5R 2M2

(403)484-3187

Reader Service Number 203

UPPER DECK FORTH $49 >

Based on Forth-83 Standard Fully segmented
architecture Fast compile, no link required
Uses standard DOS files Integrated multi-file full
screen editor Assembler, decompiler,
source-level debugger Turnkey application
support For IBM PC/XT/AT and compatibles
with 256K, hard disk or floppy, DOS 2.0 or later
Add $3 for shipping and handling (outside
USA $ 15). CA residents add sales tax.

Upper Deck Systems
P.O. Box 263342,
Escondido.CA 92026 (619)741-1075 ,

< S

Reader Service Number 211

Presto Print

A utility that sends print files to disk fast
Lets you work and print simultaneously

1 Uses leas than 6k • Uses any LPT/COM port
' Provides printer selection • Supports Xon/off
■ Auto-releases disk space when file Is printed
' You select amount ot disk space to be used
' Doesn't Interfere with computer usage

Only $49" 1-800-274-3007
*V Send check or money order to:

Gribbs & Associates
331 Glencrest Or. Solana Beach CA 92075

Reader Service Number 214

sPORTt

serial PORT tester
sPORTt is a serial port tester for people who don't want to
be data communications experts, but use or write software
for serial devices, such as plotters, mice, digitizers,
modems, serial printers, or computer to computer links.

• displays all data going in and out

• displays ASCII, decimal, hex, binary

• lets you choose handshaking lines

• sends/receives disk files

• automatically runs port diagnostics

Order today only $80

WHS

3037 Grass Valley Hwy #8201 Aubum, CA 95603
916-885-2480

Reader Service Number 192

16 Megabytes ems and/or

Extended Memory

XT and AT
Compatible

•Works on 8 or 16 bit bus
•16 bit transfer on AT bus
•Single board design
•Includes RAM disk and
extensive diagnostics
•Quantity/OEM discounts

Designed,
Manufactured,
Sold and Serviced by

007 North 6th 8t Uric* City, MN 65041 (612)345-4555

Reader Service Number 3

LOW-BAND PC LOGIC ANALYZER

• Designed for general applications (below 1 MHz)
where high-cost analysis is not needed

• Works with 5V digital logic

• 8 color-coded data lines & external clock input

• 6 range internal clock (1 Hz-1 OOKHz) for periodic sampling

• 8-bit logic probe mode

• Card plugs into IBM PC/XT/AT/386

¥-uMuuuruiniuu — in
Lniuinnnn_iun_ruiJinjuu
iMannLJUiJinjLJUTJiru~iri
iruinoiLiJimji nruruu

36" cable with micro-clips

Full-featured software

Microsoft-mouse compatible

Fast assembler data acquisition

Help system & manual

PHOTRONICSl $99.95

109 Camille Street MC.VISA,

Amite, LA 70422 Company PO,

504-748-7090 C0D

Reader Service Number 189

FLOPPY DISK FILE

Stackable - For Convenient Storage

Adjustable Metal Follower Keeps Disks Upright
and Neat

Sturdy Fiberboard
Construction

Easy Pullout Drawer
With Metal Brasstone
Handle and Card Holder

10 Dividers Included

Available for 5.25" and
3.5" Floppy Disks

Disk Storage
Model Size Capacity

33-1 3.50" 100
55-1 5.25" 150

Price Size (HxWxL) In.

$15.70 5.25x4.25x15
$17.20 7x6.5x15

Plus shipping $3.20 Per Unit - In Florida, Add 6% Sales Tax

**No shipping charges with orders of 12 or more boxes**

WE ACCEPT VISA AND MASTERCARD

L0ALC0 Inc.

P.O. Box 7117, Delray Beach, Florida 33484
Telephone: 800-869-9241 • (407) 243-6222

Reader Service Number 215

Why you want BATCOM!

BATCOM is a batch file compiler that compiles
your *.bat" files to \exe" files to make them faster,
more professional, and more capable. BAT-
COM extends DOS with new commands so you
can read keyboard input, perform arithmetic, use
subroutines, and much more. In addition, BAT-
COM protects your source code, and you can
distrbute your compiled programs without royal-
ties. For IBM PC. Only $59.95. Order todayl
y~*\ Wenham Software Company
f • | 5 Burley St.
\_S Wenham, Ma. 01984
(508) 774-7036

Reader Service Number 124
MICRO CORNUCOPIA, #53, May, 1990 93

NEW COMPUTER BOOKS

Select from thousands of new
titles in C++, C, ADA, Pascal,
Modula, Assembler, Basic, Lisp,
Fortran for PC's, MAC, Unix and
VAX. CAD, Desktop Publishing,
Dbase, Word Processing,
Business & more. Not a book
club, no membership fee or
required purchases. Service for
businesses & professionals.

Quality at low prices.

up to 40% OFF LIST

SMALL MINIMUM ORDER

send for FREE information to

Computer Book Co.

PO 7076

Sunnyvale, CA

94086-0746

Reader Service Number 167

'Coming Up In Issue #54 ^

The meaning of life, the universe and
everything. This series will take a quantitative
look at this engrossing subject rather than
mucking about in silly philosophical mud
puddles.

You won't find this issue at your local
newsstand, nor will it be available by
subscription. Issue #54, coming soon to a place

Very Close to you. Watch for it. (Thanks, Dave Stojan.)

SHAREWARE AUTHORS: Give your
programs the chance they deserve! With
Megapost, your program is uploaded to
CompuServe, GEnie, Delphi, BIX, and
EXEC-PC. The low fee is probably less than
you would pay just to open accounts with all
these services. Included are one half-price
update and notification of the downloading
procedure and 3-month download count for
each service. 15% discount for ASP
members. Write for a free brochure.

Andrew M. Saucci, Jr.
641 Koelbel Ct. Baldwin, NY 11510-3915

Reader Service Number 200

Micro
Ads

Figure 2 —

STKSIZ.ASM

.MODEL

SMALL

EXTRN

_end:WORD

; stack bottom

EXTRN

_ _amsg_exit : PROC

; write message and exit

.DATA

PUBLIC

STKHQQ

offset DGROUP:_end+256 ; stack bottom + slop

PUBLIC

stksiz

dw
.CODE

-1

; larger than possible

PUBLIC

chkstk

_ _chkstk

LABEL

PROC

pop

; return offset

@codesize

pop

; return segment

ENDIF

mov

bx, sp

sub

bx, ax ;

subtract local requirement

overflow

; stack overflow

cmp

bx, {STKHQQ]

overflow

; stack overflow

mov

sp,bx

; new stack pointer

sub

bx, [STKHQQ]

; get free stack space

cmp

bx, [ stksiz]

jae

return

mov

[ stksiz] ,bx ;

replace with smaller value

return :

Gcodesize

push

; return segment

push

; return offset

dummy

PROC

FAR

ret

; return to DX:CX

dummy

ENDP

ELSE

jmp

; return to CX

ENDIF

overflow:

xor

ax, ax

jmp

_ _amsg_exit

; write message and exit

END

♦ ♦

♦

"stack probe" routine (called at every
function entry point to verify that there
is enough remaining stack space to allo-
cate local variables required by the func-
tion) could be modified to also maintain
a global variable that would contain the
smallest amount of free stack space up to
that point.

The routine presented here (see Fig-
ure 2) was written for Microsoft C and is
functionally equivalent to the Microsoft
stack probe routine (called _chkstk) ex-
cept that a global variable called _stksiz
is maintained. My routine is only a little
larger and slower than its counterpart.

To use my routine you must first de-
lete the chkstk module from the Micro-
soft run-time library you are using. This

is because chkstk is not a procedure

label.

When testing a program, I print

_stksiz just before exiting. Assuming
stack checking is enabled (see your com-
piler documentation), _stksiz will con-
tain the smallest amount of free stack
space for the portion of the program that
was actually executed.

The routine presented here is for the
small memory model and was written
for MASM 5.0. To change memory mod-
els you need only change the .MODEL
directive.

Jack Whitney
Whitney Software, Inc.
P.O. Box 4999
Walnut Creek, CA 94596
(415) 933-9019

♦ ♦ ♦

94 MICRO CORNUCOPIA, #53, May, 1990

72 Acquired Intelligence 83

187 Alphax Systems, Inc 78

160 Annabooks 76

4 Austin Codeworks 31

197 Az-Tech Software, Inc 93

147 Berry Computer 49

214 Bluebird Systems 93

31 CCSoftware 80

185 Carlson 91

15 Cascade Electronics 39

217 Catenary Systems 75

167 Computer Book Co 94

7 Compuview Products, Inc 11

206 Coronado Enterprises 91

158 Cottage Resources 91

10 Emerald Micro ware 29

93 EracCo 45

204 Gateway Technologies, Inc 91

212 General Software 63

** Genus Microprogramming 19

149 Information Modes 7,91

22 Integrand 6

213 JB Software 5

198 Legend Communications 93

215 Loalco 93

DVERTISER'S INDE

ISSUE 53

181 MCR 91

** Micro Cornucopia 74,77

37 Microprocessors Unlimited 63

2 Microsphere, Inc. ... 94, Inside Front
** Midnight Engineering 87

186 NOHAU 93

110 Nu-MegaTech 2

201 OPENetwork 69

193 Oregon Software 75

216 OSCS 53

161 Opal Fire Software 79

3 PCTech 93, Back Cover

113 Paradigm Systems 80

119 Peripheral Technology 78

189 Photronics 93

** Programmer's Journal . . .Inside Back

139 QuantasmCorp 72

** RJSwanteklnc 37

210 Regan Consulting & Programming . 66

** Revolution2 25

166 Robotic Systems 79

171 RyleDesign 84

176 Sampson Engineering 91

200 Saucci 94

127 SemWare 13

162 Semi-Disk Systems 84

** Sintar Software 76

219 Software Blacksmiths, Inc 56

129 Software Science, Inc 33

40 Star-K Software Systems 18

202 Symbologic 59

182 Symmetric Research 91

** Turbo Users Group 87

218 TechnicaRex 91

178 TraxelLabs 32

194 Turbo Power 51

211 Upper Deck Systems 93

62 V Communications 43, 82

208 Vernier Software 44

124 Wenham Software Co 93

192 Western Hydrologic Systems 93

169 Western Wares 72

164 Whitney Software 91

203 Winter City Software 93

39 Xenosoft 93

207 Young Software Engineering 23

70 Zortech 1

** Contact Advertiser Directly.
When you write for information, please tell these folks you read about their products in Micro Cornucopia.

What's this I hear, no more Micro C?

No more around the Bend or SOG?

What will become of the human race

("All watched over by machines of loving grace")

With no PC mags of culture and taste

to keep computers in their place?

Whatever the reason, I know it's just.

Close down the rag if you think you must.

But know your humor will be sorely missed —

If you start up another put me on your list.

(sob)

Old friend, I hardly knew ye.

(Quotation courtesy Richard Brautigan)

-Buddy McManus

MICRO CORNUCOPIA, #53, May, 1990 95

Man-Machine

By Gary Entsminger

P.O. Box 2091

Davis, C A 95617

THE

LASTPAGE

Interactions

The Edges of Reality

Can you control your own reality — even
a little bit? Here's a way to find out.

Thanks in part to the "strange"
point of view cast by quantum
physics in this century, philoso-
phers, psychologists, other-ologists, and
just plain folk have readdressed the clas-
sic question: What role does consciousness
play in the creation of reality?

In particular, the Princeton Engineer-
ing Anomalies Research Program
(PEAR), has been challenging the com-
mon sense view of reality through care-
fully controlled experiments.

Robert Jahn and Brenda Dunne have
collected ten years' worth of PEAR re-
sults in a fascinating book, Margins of Re-
ality, the Role of Consciousness in the
Physical World. Margins attempts to:

(1) produce a history of scientific at-
tempts to credit and discredit paranor-
mal events;

(2) define parapsychology;

(3) review the design, operation, and
results of an ensemble of engineering ex-
periments. These experiments address
the interaction of consciousness with
both inanimate devices and systems
embodying random processes;

(4) create a theoretical framework to
explain the PEAR results.

The story Margins relates is, to say the
least, ver-y in-ter-est-ing.

Man-Machine Connections

The PEAR Program's experiments fall
into two broad categories:

(1) man-machine interactions;

(2) precognitive remote perception.

PEAR's interest in man-machine in-
teractions stems from concern about the
sensitivity of technology. One possible

subtle influence is human consciousness.

I found two of the PEAR experiments
especially intriguing. Both utilize ran-
dom event generators (REGs) to create
random patterns. One uses microelec-
tronic circuitry; the other uses a random
cascade apparatus (a macroscopic con-
traption, 10 feet wide and 6 feet high,
which lets 3 A" balls funnel down
through a quincunx array of 330 pegs).

In both experiments, PEAR estab-
lishes a control by running the REGs un-
attended. These unattended runs
establish the likelihood of random out-
put. Then an operator positions herself
several feet from the REGs and tries to
affect the output by thinking about it.

In the microelectronic experiment, the
experiment simulates a coin toss (the
outcome is binary: 1 or 0). The operator
tries to increase the output of either Is or
0s (in the controlled runs the output of
Is and 0s is roughly equal).

In the macroscopic experiment, the
controlled runs produce a cascade of
balls that's bell-shaped. In this scenario
the operator now tries to shift the cas-
cading of balls either left or right.

The results, as you might expect, vary
among operators. Some had an extraor-
dinary ability to affect the output, partic-
ularly as the number of experimental
runs increased.

Typically, an operator's effect wasn't
significant in the first few hundred or so
runs. But as the number of runs in-
creased, the operators got better.

My Test/Your Test

During the past 15 years, I've en-
countered many paranormal claims,
most of them seriously flawed one way
or another. Most often, experimenters
claim more than their experiments show,
fail to completely isolate variables, or are

unable to replicate their results.

PEAR avoids the common pitfalls of
paranormal research, doesn't claim too
much, and seems well on track to dis-
covering important clues to the role con-
sciousness plays in creating reality.

Just for fun, try an experiment. Set up
an REG using the randomize and ran-
dom functions in Turbo Pascal or C.
Make sure you call randomize before
each call to random, otherwise each run
is really only one random event and seri-
ously flawed as a control for the experi-
ment. Calling randomize reshuffles the
deck by calling the system clock to
reseed the random number generator.
For example —

uses CRT;
yar

I, R, Totl, Tot 0: integer; C: char;
begin

Totl := 0; TotO := 0;
for I := 1 to 10000 do begin
randomize; R := random (2);
if R = 1 then Totl := Totl + 1
else TotO := TotO + 1;
end;

writeln ( ' Totall : ' , Totl) ;
writeln ( ' TotalO : ' , TotO) ;
end.

Generate ten groups of ten thousand
samples. Call it the control. Then repeat
the experiment, but this time concentrate
throughout the process (an event at a
time) on either a 1 or outcome. Who
knows — maybe the consciousness of
Micro C affects reality?

References

Jahn, Robert G. & Brenda J. Dunne;
Margins of Reality; Harcourt, Brace,
Jovanovich; 1987; $17.95 (paperback).
♦ ♦ ♦

96 MICRO CORNUCOPIA, #53, May, 1 990

The complete Programmer's Journal "On Graphics" series-assembled in one volume for the very first time!

Maximize your EGA/VGA programming
graphics with Que's Power Graphics
Programming! Que has teamed up with
Programmer's Journal to bring you
this nuts-and-bolts graphics pro-
gramming guide. Now, for the first
time, you can have the expertise
of Michael Abrash consolidated
in a single volume. His insights
and approach to efficient,
effective programming have
earned him high acclaim for
his "On Graphics" series.
\ If you are seeking that elu-
sive graphic edge, you
' can't be without Power
Graphics Programming*.

Order yours
directly from
Programmer's Journal

To order call 800-234-0386.
The Power Graphics Programming
code disk is also available for $14.95.
Order your book and code disk today!

Available in better bookstores and computer stores
everywhere. Or call 800-428-5331, ext. PJ01.

que

Contact Advertiser Directly

VERY HIGH PERFORMANCE

Processors, Memory, and Display Adapters

The X24 High performance processor

• 12 or 16 MHz 80286 with NO WAIT STATES !

• Small size ("XT" height and length) passive bus design

• 1 to 4 Mbyte wait state dynamic memory

• Fully "AT" compatible Award BIOS

• Runs DOS versions 2.2 and later, Xenix and OS/2

The X24 combines the best of motherboard and backplane designs in a 100% AT compatible system. Incor-
porating a 16 MHz 80286, the X24 processor is designed to operate with the PC Tech Advanced System Mother-
board, which contains the peripheral interfaces (hard disk, floppy disk, two serial ports and a parallel port). The
X24 processor can also be used with other totally passive bus backplanes. Most critical components including
the microprocessor and up to 4 megabytes of fast memory are contained on a single PC size plug-in card. This
allows the processor and main system memory to be serviced or upgraded without disturbing other peripherals
such as serial ports and disk drives.

PC Tech X24 and ASMB
The PC Tech Advanced System Motherboard

• Built in "IDE" interface for AT interface type hard drives

• Fully AT compatible floppy disk support for 3.5", 5.25" drives, capacities of 360k, 1.2m and 1.44m

• Two serial ports and one parallel port

• 8 total expansion slots PC/XT/ AT compatible (4 slots have 32 bit bus)

The PC Tech Advanced System Motherboard is designed to complement PC Tech's X24 and X32 high performance processor cards. It contains the mass storage interfaces necessary for a complete sys-
tem, plus the basic I/O required in most systems, Extra care has been given to FCC compliance by design.

34010 Monochrome Graphics Adapter II

PC Tech Mono-ll

Up to 384k bytes display memory
Up to 2 Megabytes program memory

Software is RAM based, allowing complete operating software
replacement and timing re-programming from the host bus
34010 program loader included. Assembler, debugger, and C
compiler available.

Full hardware and software CGA, MDA and Hercules emula-
tion

Single bit shared memory bit-map with optional resolution up
to 2048 x 1536 (736 x 1008 standard)

• Very high resolution COLOR version available

• Custom 34010 software development available

The TMS34010 is a true general purpose graphics processor. PC Tech makes the total processing power of the 34010 available to both programmers and end users. Our 34010 Monochrome Graphics
Adapter is designed to allow programming from the PC/XT/AT host bus. You can completely replace our 34010 software with yours to directly harness the incredible image processing power of the TMS
34010 for your application. We make a complete set of development tools available, including an assembler, C compiler, program loader, 34010 debugger, and PC interface tracer/debugger. Our stand-
ard product includes support for extended CGA, MDA and Hercules emufation as well as a host addressable graphics bit- map. We also support and recommend the DGIS graphics interface standard (from
Graphic Software Systems) for applications development as an alternative to native 34010 software development. Ready to run drivers are available for most major applications software packages as well.

Custom Designs Available

PC Tech will license most products for non-exclusive manufacture. We will also cus-
tomize any of our designs to better meet your needs on our in-house CAD systems.
All of our standard products are available in private label versions.

About PC Tech

PC Tech has been designing, manufacturing and marketing high performance PC re-
lated products for over three years. Our standard product line includes processor,
memory, and video products. All products are designed, manufactured and sup-
ported in our Lake City, Minnesota facilities.

Designed, Sold and Serviced By:

907 N. 6th St., Lake City, MN 55041
(612) 345-4555 • (612) 345-5514 (FAX)

PC, XT, AT, DGIS, Hercules, and Award BIOS are trademarks or registered trademarks of their respective companies.

High resolution fractal produced
on the PC Tech COLOR 34010

Reader Service Number 3

Internet Archive Audio

Featured

Top

Images

Featured

Top

Software

Featured

Top

Books

Featured

Top

Video

Featured

Top

Mobile Apps

Browser Extensions

Archive-It Subscription

Save Page Now

Full text of "microCornucopia :: Micro Cornucopia #53 May90"

See other formats